HOSPITAL USE AND MORTALITY AMONG MEDICARE BENEFICIARIES IN BOSTON AND NEW HAVEN

Transcription

1 SPECIAL ARTICLE HOSPITAL USE AND MORTALITY AMONG MEDICARE BENEFICIARIES IN BOSTON AND NEW HAVEN JOHN E. WENNBERG, M.D., JEAN L. FREEMAN, PH.D., ROXANNE M. SHELTON, M.A., AND THOMAS A. BUBOLZ, PH.D. From the Department of Community and Family Medicine, Dartmouth Medical School. Hanover, NH 03756, where reprint requests should be addressed to Dr. Wennberg. Supported by a grant (HS05624) from the National Center for Health Services Research. Abstract: We compared rates of hospital use and mortality in fiscal year 1985 among Medicare enrollees in Boston and New Haven, Connecticut. Adjusted rates of discharge, readmission, length of stay, and reimbursement were 47, 29, 15, and 79 percent higher, respectively, in Boston; 40 percent of Boston's deaths occurred in hospitals as compared with 32 percent of New Haven's. High-variation medical conditions (those for which there is little consensus about the need for hospitalization) accounted for most of these differences. By contrast, discharge rates for low-variation medical conditions (which tend to reflect the incidence of disease) were similar. Inpatient case-fatality rates were lower in Boston than in New Haven (RR = 0.85; 95 percent confidence interval, 0.78 to 0.92), but when all deaths (regardless of place of death) were measured, the mortality rates in Boston and New Haven were nearly identical (RR = 0.99; 95 percent confidence interval, 0.93 to 1.05). We conclude that the lower rate of hospital use by Medicare enrollees in New Haven was not associated with a higher overall mortality rate. Population-based as well as hospital-based statistics are needed to evaluate differences in hospital mortality rates for high-variation medical conditions. (N Eng J Med 1989; 321: ) PREVIOUS studies showed that residents of the Boston hospital-service area used about 4.5 beds per thousand population, as compared with less than 3 per thousand for residents of the New Haven, Connecticut, hospital-service area (1, 2) and raised questions about the possible withholding of care in New Haven (2, 3). In 1982 the total number of days in the hospital and expenditures per capita were 44 and 100 percent higher, respectively, in Boston than in New Haven. The higher rate of hospital use among Boston residents was largely due to higher discharge rates involving high-variation medical conditions, such as pneumonia, gastroenteritis, and chronic obstructive lung disease, for which there is little consensus about the need for hospitalization (2). The hospital-discharge rates involving those conditions are known to be highly variable across small geographic areas and are highly correlated with the number of available beds per capita. By contrast, discharge rates involving myocardial infarction, stroke, and gastrointestinal hemorrhage low-variation medical conditions were virtually the same in the two communities. For these low-variation conditions, which are characterized by professional consensus on the need for hospitalization, the hospitalization rates are more closely related to the incidence rates of the diseases (3, 4). What are the implications of these different patterns of use and deployment of resources? Were hospital services underused in New Haven such that the lower rates of hospitalization for high-variation medical conditions were associated with worse outcomes? Although the full exploration of these important issues requires detailed prospective studies, one question can be addressed retrospectively with data on Medicare claims. As adjusted for age, sex, and race, was the overall lower rate of hospital use associated with higher mortality rates? We show in this study that the answer to this question depends on whether mortality is viewed from a hospital-based or population-based perspective. Clinical policies regarding which patients are admitted or readmitted to the hospital, how long they stay, and whether those with terminal illnesses are commonly treated there or elsewhere differed between New Haven and Boston. These clinical policies materially affected hospital-based mortality statistics because they affected the average severity of illnesses among hospitalized patients, the average period during which those with a given condition were under observation,

2 and the likelihood that deaths would appear in hospital-based statistics. Most statistics based on hospitalassociated deaths suggest that mortality rates in Boston were lower than those in New Haven. However, if all deaths (regardless of the place of death) are counted and the rates calculated on the basis of the population, the mortality rates in Boston and New Haven were virtually the same. METHODS Study Population Our study s population consisted of all Medicare beneficiaries 65 years of age or older who were residents of the Boston and New Haven areas between October 1, 1984, and September 30, 1985 (fiscal year 1985). The Boston hospital area includes Boston, Chelsea, Revere, and Brookline, Massachusetts. The New Haven area includes New Haven, West Haven, and East Haven, Connecticut, Data Our study was based on the hospital-claims file (Medicare Part A) and the enrollment file maintained by the Health Care Financing Administration (7). The hospital-claims file contained a record of each hospitalization, clinical and financial information, and selected characteristics of the beneficiary. Elements included diagnoses (up to five); procedures (up to three); the number of the diagnosis-related group (DRG); the patient's age, race, sex, and residential ZIP Code; and the total charges and those reimbursed to Medicare. The enrollment file contained information on all the patients enrolled in the program, including their dates of birth and death, Information was linked across files to determine the number of inpatient deaths and deaths within 30 days of hospital admission and the total number of deaths. Population-Based Rates of Use Rates of use were based on hospital-discharge information from the Medicare Part A tile and population estimates from the enrollment file. We estimated the population of the two study areas at the study's midpoint, April 1, 1985, using year-end enrollment figures from calendar years 1984 and 1985 (three quarters of the December 1984 figure plus one quarter of the December 1985 figure). Unless otherwise specified, all rates in this report were adjusted for demographic characteristics by the indirect method (8). Rates of hospital use (number of discharges per capita, days in the hospital, and reimbursements) for Boston and New Haven included all the discharges of residents, regardless of where their hospitalization occurred. Types of Hospital Discharge Rates of use are also reported for specific types of hospital services in Boston and New Haven. For this purpose we divided all hospitalizations into three categories according to DRG low-variation medical conditions, high-variation medical conditions, and conditions involving surgery (as defined previously (4). Low-variation medical conditions included stroke (DRG number 14), heart attack (121 to 123), and gastrointestinal bleeding (174 and 175). High-variation medical conditions included all other medical DRGs. Hospitalizations involving surgery included all DRGs that are defined on the basis of operating room procedures. The data included 1,431 discharges (of a total of 38,525) in unclassifiable DRGs: 704 in number 468, and 727 in number 470. Those in DPG number 468 were considered surgical. Those in DRG 470 that involved a surgical procedure (286 discharges) were included in the surgical category, and those that did not (441 discharges) were included in the high-variation medical category. Mortality Rates Mortality rates based on all deaths and on hospital-associated deaths are presented for each area. We calculated the population based total mortality rate using all deaths of residents (regardless of the place of death) as the numerator and the estimated size of the Medicare population as the denominator. The number of hospitals associated deaths is reported in two ways. In the first, we counted only inpatient deaths - those in which the beneficiary died while hospitalized, as determined

3 by the date of death from the enrollment file and the date of discharge fro. the hospital-claims file. In the second, we counted deaths that occurred within 30 days of the admission date corresponding to the beneficiary's last hospital discharge in fiscal year 1985, regardless of where death occurred. We calculated the hospital-associated mortality rates using two denominators: the number of relevant discharges in fiscal year 1985 and the estimated Medicare population. The relevant discharges for the inpatient mortality rates were all discharges in the fiscal year that corresponded to the particular type of hospital-associated death (those involving high-variation medical conditions, low-variation medical conditions, or surgery). The relevant discharges for the 30day mortality rates included only the patients' last hospital dis. charges in the fiscal year. This method of computing 30-day mortality rates was used by the Health Care Financing Administration in determining hospital-associated death rates for 1986 (9). We calculated hospital-associated mortality rates on the basis of the estimated Medicare population as a result of Left and Showstack's (10) suggestion that when an adequate means of adjusting for differences in admission criteria is not available, the appropriate denominator for hospital-associated mortality statistics is an area's population rather than the number of hospitalizations. As with the rates of use, all mortality rates were indirectly adjusted (8) for age, race, and sex, and variances were computed with the method of Keyfitz (11). Ninety-five percent confidence limits were calculated for all the Boston: New Haven ratios on the basis of the corresponding variances in each area (12). RESULTS Hospitalization Rates among Medicare Beneficiaries The Medicare populations in Boston and New Haven were similar with respect to sex and race: each was about two-thirds female and almost 15 percent nonwhite. Beneficiaries in Boston were older; 48 percent were at least 75 years of age, as compared with 41 percent in New Haven. The rates of use of hospital services, as adjusted for age, sex, and race, differed substantially in the two communities (Table I)- Beneficiaries in Boston had a discharge rate 47 percent higher than that of beneficiaries in New Haven. Moreover, their hospital stays were 15 percent longer, resulting in a 68 percent larger number of days in the hospital per thousand beneficiaries. Reimbursements were 22 percent higher per case and 79 percent higher per capita. The greater overall use of hospital services per capita in Boston can be attributed to both a higher percentage of hospitalized persons (21.3 vs percent per year) and a higher rate of readmissions (32.9 vs percent per year). Larger differences in rates of use between the two communities were observed in hospitalizations associated with an inpatient death:

4 the average length of stay was 55 percent longer and reimbursements per case were 71 percent higher in Boston than in New Haven; per capita rates of hospital use and expenditures were about twice as high in Boston as in New Haven. In Boston, approximately 65 percent of all discharges involved high-variation medical conditions, 8 percent low-variation medical conditions, and 26 percent surgery. In New Haven, the corresponding figures were 59, 12, and 29 percent. Consistent with previous studies, the overall increase in hospital use in Boston was largely attributable to the provision of services to patients with high-variation medical conditions (Table 2). For these conditions, rates of discharge and number of days in the hospital were-62 and 95 percent higher in Boston. Moreover, reimbursements per capita were twice as high in Boston, and 54 percent of all inpatient reimbursements for Boston hospitalizations involved high-variation medical conditions. If the rates of use for people in Boston had been the same as those for people in New Haven, approximately 136,000 fewer days of hospitalization would have been required. Of these, 92,000 days (72 percent) involved patients with high-variation medical conditions. By contrast, discharge rates for patients with the low-variation conditions were almost identical, although Boston's lengths of stay were 13 percent longer. Likewise, for terminally ill inpatients, differences in rates of hospital use were principally related to the high-variation medical conditions: per capita rates of reimbursement and number of days in the hospital were 157 percent higher in Boston. A previous study(2) found that the higher rate of hospitalization for surgery in Boston was attributable to differences in the rate of hospital use for minor surgical procedures, Overall Population-Based Mortality Rates The overall mortality rates, as adjusted for age, race, and sex, were virtually the same in Boston and New Haven in fiscal year There were no significant differences according to age, sex, or racial group (Table 3).

5 Mortality According to Place of Death and hospital Associated Mortality Clinical preference among the clinicians who treated the Medicare population in Boston favored the inpatient setting for the care of the terminally ill. Approximately 40 percent of the deaths among Medicare beneficiaries in Boston occurred when the decedents were inpatients, as compared with 32 percent of the deaths in New Haven (Table 4). The case-fatality rate (inpatient deaths per 100 discharges) was lower for enrollees in Boston, but the higher discharge rate per 1000 enrollees resulted in the higher population-based inpatient mortality rate (1.47 X 0.85 = 1.25). Most of the inpatient deaths occurred among patients with high-variation medical conditions. Although the case fatality rate among patients with high-variation conditions was lower in Boston, the higher per capita discharge rate involving those conditions combined with the lower case-fatality rate produced a 41 percent higher population-based inpatient mortality rate for beneficiaries in Boston (1.62 X 0.87 = 1.46) By contrast, for patients with low-variation medical conditions, the discharge rate, the number of deaths per 100 discharges, and thus the per capita inpatient death rate were virtually the same in the two populations (0.99 X 1.01 = 1.00). Case-fatality rates based on the 30-day observation period were about 30 percent higher in New Haven than in Boston. However, the mortality statistics based on the number of deaths per capita within 30 days of the admission date corresponding, to the last hospital discharge were nearly the same among the two populations. For enrollees in Boston, the use of a 30-day fixed observation period had little effect on the number of deaths classified as hospital associated; the additional 343 out-of-hospital deaths that fell within the 30-day period were offset by the elimination of 313 deaths among inpatients whose lengths of stay exceeded 30 days. However, this method resulted in a 24 percent increase in the number of hospital associated deaths in New Haven, from 457 to 567. The increase occurred because 150 out-of-hospital deaths were added, whereas only 40 deaths were excluded because of stays of more than 30 days. DISCUSSION We have documented higher rates of discharge and readmission, longer stays, and higher expenditures for hospitalization among the Medicare population in Boston as compared with that in New Haven. Most of the differences in the allocation of hospital resources were accounted for by greater use of the hospital by patients with high-variation medical conditions. Hospitalization for six high-variation conditions alone - pneumonia (DRG numbers 89 and 90), heart failure and shock (127), gastroenteritis (182 and 183), diabetes (294), cardiac arrhythmia (138 and 139), and chronic obstructive lung disease (88) - accounted for 25 percent of the difference in the rate of patient days involving treatment for medical conditions. Clinical practices also differed with regard to the care of the terminally ill: 40 percent of the deaths in Boston occurred in hospitals as compared with 32 percent of those in New Haven; rates of patient days and reimbursement for terminally ill patients who had high-variation medical conditions were about 2.6 times higher in Boston than in New Haven.

6

7 Although total population-based mortality rates were similar in the two areas, hospital-associated mortality rates varied according to the measure employed. When measured as case-fatality rates (deaths per 100 discharges), hospital-associated death rates were higher in New Haven. The increase was due principally to deaths among patients with high-variation medical conditions, both inpatient case fatalities (rates 15 percent higher) and deaths occurring within 30 days of admission (rates 33 percent higher). When measured as deaths per capita, inpatient death rates were 41 percent higher in Boston. However, when measured as deaths occurring within 30 days of admission, the hospital-associated mortality per capita in the two communities was virtually the same. We concluded that these statistics did not imply differences in the skill of clinicians and hospitals in Boston and New Haven in preventing unnecessary deaths, but resulted instead from differences in clinical policies. The limitations of hospital-associated mortality statistics have been discussed extensively, particularly in response to the Health Care Financing Administration's release of hospital-associated death rates and Shortell and Hughes (13) use of such statistics in their study of the effects of regulation on patient outcomes. Although most attention has focused on case severity, the confounding effects of systematic differences in admission and readmission policies, length of stay, and place of treatment of the terminally ill are also at

8 issue. These factors differ simultaneously and variably from region to region, between communities within regions, and from hospital to hospital within communities. Intuitively, one would expect higher rates of admission for medical conditions to be associated with a lower average case severity. However, many factors may confound this association. The availability of home care services, nursing homes, and hospice programs, which affect the likelihood of a patient's dying in the hospital, varies from community to community. Also, a community's use of alternative services is not necessarily a function of the number of hospital beds per capita (14) and thus may not be related in any consistent way to admission thresholds for high-variation medical conditions. Varying lengths of stay and readmission rates add another level of complexity by affecting patients' eligibility for observation. Length of stay affects the inpatient period of observation. When hospital mortality rates are based on a 30-dav period of observation (as used by the Health Care Financing Administration), variations from institution to institution in readmission thresholds result in differential censoring and a downward bias for low variation conditions and surgical procedures in hospitals with low thresholds. For example, patients admitted with myocardial infarction who are readmitted within 30 days with a different diagnosis will not appear under "acute myocardial infarction" in the mortality statistics. Because practice patterns for high-variation medical conditions vary as much as they do, comparing the performance of health systems in treating patients with these conditions presents serious difficulties. The inconsistencies between measures that this study illustrates should be expected. At the community or regional level, population-based statistics that include the total number of deaths as well as hospital-associated deaths can be collected, and the effects of practice patterns can be analyzed with population-based data on use. Errors of interpretation attributable to the confounding effects of practice style can thus be avoided. However, the population at risk for admission to any one hospital is not ordinarily observable. Thus, admission and readmission rates, lengths of stay, and the effects of policies concerning the care of the terminally ill cannot be directly measured for individual hospitals. For this reason, variations in hospital-based mortality rates among patients with high-variation medical conditions (15, 16) should not be interpreted as reflecting differences in clinical skill or productivity. What does our study say about the productivity of hospitals? If the illness rates were similar in Boston and New Haven, there was no discernible difference in survival associated with an 80 percent difference (adjusted for age, sex, and race) in Medicare reimbursements. But what is the evidence that illness rates were similar, given the large difference in rates of hospital use? First, it should be rioted that rates of use for high-variation conditions provide no information on relative illness rates: the discharge rate among patients with high-variation medical conditions is highly correlated with the number of hospital beds per capita but not with illness rates (6). Second, socioeconomic factors were similar in Boston and New Haven (2). Third, the similarity in rates of' discharge involving acute myocardial id1rction, stroke, and gastrointestinal hemorrhage - the low-variation conditions about which physicians agree on the need for hospitalization - is a direct indication that illness rates, at least for common illnesses, such as coronary artery and cerebrovascular disease, were more or less the same. Moreover, the similar mortality rates among patients with these conditions suggest similar end results, at least in one important dimension. Finally, there is no evidence that an increased number of high-variation medical admissions leads to lower mortality. Despite the differences in use and deployment of resources for high-variation medical conditions and the apparent similarity in illness rates, there was no difference in overall population based mortality rates among Medicare beneficiaries in Boston and New Haven. Our study thus supports the opinion of clinicians in New Haven that the lower rate of hospital use in treating high-variation conditions did not constitute a withholding of valuable services (2).But mortality is only one measure of outcome. Did the higher rates of hospitalization in Boston result in less morbidity, fewer complications, and an improved quality of life? Was the dying patient's quality of life better in Boston because hospitals were used more frequently to treat the terminally ill? Beyond the availability of hospital beds, what structural characteristics in the medical care systems of the two cities promoted their differences in practice style? These questions remain. Because the answers may create opportunities to reduce the cost of medical care without damaging the welfare of patients, studying different approaches to the treatment of high-variation medical conditions should have the highest priority oil our research agenda.

9 REFERENCES 1.Wennberg JE. Dealing with medical practice variations: a proposal for action. Health Affairs (Millwood) 1984; 3(2): Wennberg JE, Freeman JL, Culp WJ. Are hospital services rationed in New Haven or over-utilized in Boston? Lancet 1987; 1: Wennberg JE. Small area analysis and the medical cue outcome problem. In: Sechrest L, Bunker J, Perrin E, ads. Improving methods in non-experimental research. Beverly Hills. Calif.: Sage Publications (in press). 4. Wennberg JE, McPherson K, Caper P. Will payment based on diagnosis-related groups control hospital costs? N East I Med 1994; 311: Roos NP, Wennberg JE, McPherson K. Using diagnosis-related groups for studying variations in hospital admissions. Health Cam: Rome Rev 19H8; 9(4) Wennberg JE. Population illness rams do not explain population hospitalization rates: a comment on Mark Blumberg's thesis that morbidity adjustments are needed to interpret small area variations. Med Care 1987; 25: Lave J, Dobson A, Walton C. The potential use of Health Cue Financing data sets for health services research. Health Care Financing Rev 1983; 5:1: Fleiss JI. Statistical methods for rates and proportions. New York- John Wiley, Department of Health and Human Services. Medicare, hospital mortality information Washington, D.C.: Government Printing Office, Lau HS, Showstack JA. Effects of regulation, competition, and ownership on mortality rates among hospital inpatients. N Engl Jour Med 1988; 319: Keyfitz N. Sampling variance of standardized mortality ones. Hum Biol1966; 38: Freeman D. Applied categorical data analysis. New York: Marcel Dekker, 1987: Stencil SM, Hughes FFX. The effects of regulation, competition, and ownership on mortality rates among hospital inpatients. N Eng J Med 1988; 31& Wennberg J, Gittelsohn A. Small area variations in health care delivery. Science 1973; 182: Daley 3, Jencks 5, Draper D, Lenhan C, Thomas N, Walker J. Predicting hospital-associated mortality for Medicare patients: a method for patients with stroke, pneumonia, acute myocardial infarction, and congestive bean failure. JAMA 1988; 260:3617~ Kahn KIL, Brook RH, Draper D, et al. Interpreting hospital mortality data: how can we proceed? JAMA 1989; 260: