Printer Friendly

Medicare use in the last ninety days of life.

This article examines changes in Medicare use during the last 90 days of life and focuses on the role of the prospective payment system (PPS) in the changes. Introduced in late 1983, the PPS was intended to encourage hospital administrators to find ways to economized on spending for the hospital stays of Medicare beneficiaries. At the same time, the Health Care Financing Administration implemented the professional review organization (PRO) program, which aimed to reduce unnecessary admissions and to control potential untoward activities under the PPS, such as unnecessary readmissions and dangerously shortened stays. These events, along with some important technological changes, have led to reduced admission rates, shortened stays, and increased use of home care, skilled nursing homes, and a variety of other outpatient services (Prospective Payment Assessment Commission 1988; Gornick and Hall 1988; Schmitz 1989).

Patterns of care during episodes of illness provide the best opportunity to examine physician responses to various pressures to make less use of the hospital. Here we investigated another "type" of episode: the last 90 days of life. Although this concept does not actually represent an episode of illness, it is instructive because of the intense use of health services associated with it (McCall 1984; Riley et al. 1986; Garfinkel 1988) and because of the emphasis policymakers place on the adequacy of Medicare when beneficiaries may need it most.

Few published studies on the nature of death episodes are available. A study on the use and costs of Medicare services in the last two years of life (Lubitz and Prihoda 1984) reports that decedents comprised 5.9 percent of their study group but accounted for 28 percent of Medicare expenditures; moreover, the study finds that 46 percent of costs in the last year of life are spent in the last 60 days. Work on the last year of life on Medicare beneficiaries (Riley et al. 1987) indicates that the cause of death is an important determinant of reimbursement levels. Studies of cancer patients (Greer, Mor, Morris, et al. 1986; Kidder, Merrell, and Dohan 1988) point to strong organizational influences on the use of home and inpatient services in the 50-day period preceding death. For example, in the Greer study, hospice patients receiving home care spent about 8 of their last 50 days in hospitals and received about 20 home care visits; hospital-based hospice patients had about twice as much hospital care and half as many home visits. Patients not enrolled in hospice care spent 23 of their last days in the hospital and received about 4 home care visits. Clearly, patterns of care for terminally ill patients are amenable to considerable substitution, and these patterns appear to be readily adaptive to organizational incentives.

In view of general trends in length of stay and admission rates, a main objective of this article is to describe corollary trends in the use of hospital services during the last 90 days of life. Have trends in general admitting and discharge practices led to lower admission rates and days of care in the "death episode"? Have the pressures to reduce length of stay changed the patterns of use of the hospital as the place of death? Secondarily, we are concerned about the possible substitutions for hospital care that may be occuring in the death episode. If less hospital care is being used, are more home health, nursing home, or professional office services being used? And, overall, have changed patterns of Medicare benefit use altered total reimbursemene levels for the last 90 days of life?

In examining these questions, we first explore recent trends in measures relating to hospital use, hospital mortality, and non-hospital service use. We then use multivariate methods to test simple hypotheses about the temporal patterns of changes before and after the introduction of the Medicare PPS.


A simple random sample of 8,000 deaths per year for the 1982-1986 period was drawn from the 5 percent Medicare eligibility file. The criteria for inclusion were a valid indicator of the date of death, Part A entitlement during the year of death, no indication of enrollment in a Medicare hospice program or health maintenance organization in the year of death, and residence in Washington, D.C. or one of the contiguous states.

Data on the use of Part A and Part B Medicare services during the 90-day period before death were obtained in each case from paid stay record and billing files. The files contain data on three types of Part A services: inpatient hospitalization (from the MEDPAR files), use of a skilled nursing facility, and use of home health care. The data on Part B services included use of the hospital outpatient department, durable medical equipment (DME), physician inpatient services, and physician ambulatory services. Area variables describing the health care field as well as population characteristics based on county of residence were also amended to the episode files; beneficiary demographic characteristics from enrollment records (date of death, age, race, sex, and location codes) were also included.

Dates of admission and discharge from the hospital are recorded precisely, allowing us to note whether or not a patient died in the hospital. Unfortunately, the files do not indicate where deaths outside the hospital occured. We know whether the beneficiary used a covered skilled nursing facility (SNF), home health care, or Part B service in the period before death, but we do not know the exact dates of admission, discharge, or services, and thus cannot determine whether the patient was under active care at the time of death. Therefore, we divided the sample into five mutually exclusive and hierarchical groups: (1) beneficiaries who died in the hospital; if not, then (2) other beneficiaries whose deaths were preceded by a skilled nursing facility stay and whose SNF stay occured after the last hospital discharge in the file; (3) persons who used the services of a home health agency (HHA) within 90 days of death but were not included in the two groups just described; (4) of the remainder, persons who had evidence of Part B or outpatient care; and (5) other persons who had no record of use of covered Medicare services within 90 days of death.

In three states--Kentucky, Ohio, and Pennsylvania--substantial fractions of bills for impatient of home health care were unavailable (the bills had been returned to the intermediary for recording and had not been reentered in HCFA's statistical file system); cases from these states were dropped from the analysis. In addition, approximately 100 cases with unreconciled inpatient bills more than 30 days after the recorded date of death were dropped. A total of 34,576 cases were analyzed.


The descriptive analysis examines the study population, trends in hospital use patterns, and differences between use trends in PPS states and waiver states. Table 1 shows the demographic features of the Medicare beneficiaries we studied. A gradual increse in longevity was evident across the short period of the study. The average age at death rose from 77.9 in 1982 to 78.5 in 1986. This trend was most pronounced among women, persons in rural areas, persons with end-stage renal disease, and disabled persons. There were smaller improvements in longevity in the other segments of the Medicare population. Table 1 also suggests the effects of an influenza epidemic in 1985. The trend toward increased longevity was slightly reversed that year, and average ages at death fell in all groups except patients with end-stage renal disease (ESRD).

The ESRD anomaly, and other changes in the age-at-death trends also reflect systematic changes in the ages of covered persons. While the "aged" group is large and stable demographically, one can see from the trend in the group with end-stage renal disease that efforts to


extend benefits for patients with this disorder to clinically "marginal" persons tended to increase the average age of the recipient population. The disabled group has also changed; efforts to limit the eligibility of the "working disabled" have increased the average of eligible persons.



Marked changes in the use of Medicare coverage services over the 1982-1986 period decreased the proportion of deaths occurring in all hospitals from 51.1 percent in 1982 to 45.4 percent in 1986 (Table 2). An increase occurred in the proportion of persons who used home care and other Medicare-covered services in the last 90 days of life. In the case of home care, the trend was evident throughout the period, no doubt the reflection of an extensive liberalization of the HHA benefit in 1981. This included, among other changes, dropping the requirement



that persons be eligible for home care only if they were discharged from a hospital (Williams, Cella, and Gaumer 1984). There has also been a decrease in the proportion of cases that used no Medicare-covered services. This latter trend may reflect an improvement in access to these services during the last 90 days of life.

Table 3 shows trends in the fractions of beneficiaries who die in hospitals in states with and without PPS and in various segments of the Medicare population. The reduction in frequency of death in the hospital is largest among white subjects, rural subjects, and those 75 to 84 years old. Clear differences in the levels of in-hospital deaths were associated with race and age; beneficiaries who were white and those over 85 were less likely to die in the hospital. Differences between urban and rural beneficiaries were no longer present by 1986.

The observed reduction in the fraction of deaths occurring in the hospital appears large (about 10 percent) for such a short period. To give a better idea of the trend, Figure 1 shows the quarterly rates of in-hospital deaths (as a fraction of all deaths among Medicare beneficiaries) in states with and without PPS. In 1982 and 1983, the fraction of in-hospital deaths was roughly 10 percent higher in the waivered states. The gap clearly widened during the next few years, as in-hospital mortality rapidly declined in states under PPS. Some decline was also evident in the waiver states. These patterns may reflect the more rapid shortening of length of stay in states under PPS. (1),(2)

Table 4 indicates that the decreases in the frquency of deaths occurring in hospitals was accompanied by slightly lower admission rates, shorter stays, and lower rates of death before discharge. Admission rates did not change very much; in 1986, about two-thirds of Medicare benefiaries were still admitted at least once during the last 90 days of life. Days of hospital care, on the other hand, fell 8-10 percent over the five years of the study; the reduction occurred between 1983 and 1985.

Data on the hospital stays of these patients indicated a substantial reduction in the fraction who died before discharge and an increase in the fraction who died within 15 days after discharge (Table 4). Some increase also occurred in the rate of death during the period 15 to 30 days after discharge, and a slight increase occurred in the death rate within 15 days following admission. These data suggest an increase in the severity of illness at the time of admission and an increase in the numbers of terminally ill patients being discharged earlier, to die outside the hospital within several weeks. Although programwide hospital admission rates for Medicare fell markedly over the period of the study (from about 381 per 1,000 persons in 1982 to about 310 per 1,000 in 1986), it does not seem that benefiaries had much less access to hospitals (lower admissions) during the last 90 days of life.



Use of the home care (HHA) benefit during the last 90 days of life rose markedly over the five years of the study; the fraction of cases using HHA services rose from 13.9 percent to 18.8 percent (Table 5). (3) Home care use during the 90-day period also rose among persons who were



hospitalized, suggesting a general increase in use; the fraction of hospital discharges receiving HHA care was up from 17 percent to 23 percent. Use of skilled nursing facilities was subject to an increasing trend, although it was less marked than the trend to home care services. There was an actual drop in use rates in the last year of the study. About 6 percent of subjects used a skilled nursing facility during the last 90 days of life. As with the use of home care, use of this benefit increased among persons who were also hospitalized during the last 90 days of life and for persons in PPS states.

The frequency of death preceded by no Medicare service use may be a market for poor access. (4) Rates of death preceded by no Part A or Part B services during the last 90 days have generally declines (Table 6), but this apparent increase in access was evident only in states under PPS. In these states, the fraction of patients dying with no billed Medicare services dropped from 9.6 to 8.2 percent, whereas it remained at about 8.0 percent in states without PPS. The trend was most pronounced among nonwhites (the fraction fell from 12.0 to 8.9 percent in this group, as compared with 9.1 to 8.2 percent among whites) and among benefiaries over 85. Interestingly, women were much more likely than men to have some billed service before death; in 1986, 10.0 percent of the men received no services, as compared with 6.2 percent of the women. Obviously, age is an important factor in explaining variations in this measure; persons under 65 are three to four times as likely to die with no services in the 90-day period than are persons over age 75.
 Table 6: Death with No Indications of Prior Medicare Use
within 90-Day Period (Percentof Cases)
 1982 1983 1984 1985 1986
Total--All subjects 9.3 9.6 9.1 8.0 8.2
PPS status
 States not under PPS 8.0 7.8 7.6 7.4 8.0
 States under PPS 9.6 10.0 9.5 8.1 8.2
 White 9.1 9.4 8.9 8.0 8.2
 Nonwite 12.0 11.9 11.4 8.2 8.9
 <65 23.7 20.5 22.5 20.1 18.5
 65-75 10.1 11.9 11.3 10.0 11.9
 75-85 6.5 6.1 6.7 5.7 5.2
 >85 8.9 9.2 7.3 6.4 6.2
 Male 11.4 11.3 12.0 10.1 10.1
 Female 7.2 7.7 6.3 5.9 6.2


We used economometric analysis to test hypotheses about the temporal pattern of trends in Medicare use during the last 90 days of life, using covariates to adjust for personal and regional differences in the sample. A maximum likelihood logit technique was used to estimate the three models with discrete dependent variables. Simple OLS (ordinary least squares) regression models were fit on other dependent measures. All dollar measures were adjusted to 1979 dollars using the Data Resource, Inc. (DRI) Medical Care price index.

We selected several covariates that are generally thought to influence the use of health care services. The first group included variables determining the availability of health care services in the region under study (the number of physicians per 100,000 population, the percentage of physicians who are specialists, the number of hospital beds per 100,000 population, and the number of nursing home beds per 100,000 population).

Several variables pertaining to the study subjects (age, race, sex, and Medicare eligibility status) and county-level demographic variables were included. The county measures were the percentage of population white, average education level, per capita income, urban/rural location, percentage of population on Part A Medicare, and AAPCC rate (a measure of previous cross-sectional differences in the use and payment levels of the Medicare program). Since the data set contained both hospitalized and nonhospitalized Medicare beneficiaries, no hospital-level descriptors were used (such as ownership or the data each hospital actually phased into the PPS system). Use of skilled nursing facilities and home care services within the last 90 days of life were also included in the regression on inpatient days in order to test whether these services were used to substitute for inpatient days.

The effect of PPS on the trends was estimated using a four-way design. This specification includes a PPS dummy variable (1 in 1984 and after, for cases from PPS states), a non-PPS state dummy variable (1 in all years if the case is from a non-PPS state), and a set of time (year) dummy variables to capture trends. It was necessary to assume that in each state where PPS was introduced, the system was implemented at the beginning of calendar year 1984. The PPS dummy coefficient measures the pre-post change in states under PPS as differenced against the pre-post change for cases from the waiver states. The concept is:

PPS effect = ([] - [PPS.sup.pre]) - ([] - [non-PPS.sub.pre])


([non-PPS.sub.pre] - [PPS.sub.pre]) is given by the coefficient on the non-PPS state dummy.

This specification of PPS effects is not without design threats. There have been significant trends in medical practice patterns during the 1980s stemming from new technology, widespread introduction of utilization review and payment safeguard programs, the new hospice benefit and Medicare, liberalization of the home health benefit, and other occurrences. To the extent that these temporal threats affect all states, the four-way design works to isolate only differential trends associated with the PPS states. But since the waivered states are few in number, are isolated for the East, and have longstanding and unique programs of hospital regulation, it may well be that the health care system in those states reacts to the sources of temporal changes in different ways. If so, such differences will be incorrectly measured and attributed as a PPS effect. Consequently, we interpret these econometric results with caution.


Table 7 lists PPS coefficient estimates and their p-values for all of the models. Parameter estimates appear in Table 8. As Table 8 shows, the number of inpatient days during the last 90 days of life decreased sharply after PPS was implemented relative to trends in non-PPS


states; we estimated a reduction of about 1.6 days of inpatient care (13 percent) in the 90-day period before death (p = .0001). The introduction of PPS also appears to be associated with a lowered probability of death in the hospital (p = .009).

Admission rates also fell after PPS, relative to those of waiver states, but not significantly so (p = .41). Consequently, we conclude that the relative reduction in deaths occurring in hospitals in PPS states was probably associated with PPS, and that reduction was probably due to shortened lengths of stay rather than reduced admission rates. Figure 1 corroborates the attribution of reduced probability of death in the hospital to occurrences in PPS states rather than to some occurrence in waiver states.

Most of the PPS coefficient estimates in reimbursement regressions turned out to be insignificant. Changes in total spending, spending for inpatient care, and spending for skilled nursing facilities were not associated with the introduction of PPS. But spending for home care increased after PPS by about 28 percent (relative to the trend in waiver states). We suspect that the capacity constraints on expanded


use of skilled nursing facilities were more severe than those on home care services, resulting in less measured substitution between skilled nursing facilities and hospital care over the short observational period. And the home care industry in waivered states (particularly New York) has been better developed and more heavily used (see Table 5); hence, we may be observing some catch-up in PPS states.

The level of Part B activities seems to have been influenced by PPS, suggesting substitution of outpatient for inpatient care. Physician office reimbursements rose about 20 percent (p = .004) after PPS was introduced (relative to waiver state trends). Total Part B payments and inpatient Part B reimbursements (mainly physician payments) had fairly large positive coefficients (4 to 6 percent of the respective means) but were not statistically significant.

In this analysis, the introduction of PPS did not appear to alter the likelihood that persons would receive some (or no) Medicare-covered service during the last 90 days of life; the logit coefficients suggest a reduction in the fraction not using services, but the standard errors are large.

The models (Table 7) show some important baseline differences between states where PPS was waived and states where it was implemented; the coefficient for the waivered state dummy captures these differences. Hospital admission rates were similar, but after adjustments, states where PPS was waived had more days of hospital care and offered a greater likelihood of death in a hospital. These states also had lower baseline levels of reimbursement for skilled nursing facilities and higher levels of reimbursement for care in a physician's office. This pattern is consistent with other studies showing longer lengths of stay and higher inpatient mortality in New York and New Jersey, where shortages of nursing home beds are pervasive (Gaumer et al. 1987).

Models also show that the availability of health services has an important influence on use and spending in the last 90 days of life. In the hospital death and admission rate models there are strong supply influences. Hospital bed availability increases both the likelihood of admission, and the likelihood of death occurring in the hospital. The signs on measures of physician availability and nursing home bed availability suggest substitution for hospital care, showing that higher levels of resources are associated with a reduced likelihood of admission and less chance of death occurring in the hospital.

The availability of hospital beds tended to be positively associated with days of care (as found by Ginsburg and Koretz 1983), but negatively associated with most reimbursement measures. This would suggest the substitution between hospital days and SNF care, home care physicians' services, and other services. Interestingly, inpatient payments were also lower when more beds were available, suggesting competitive influences rather than supply inducement. More physicians per capita, on the other hand, were associated with lower admission rates, fewer days of care, lower in-hospital death rates, and higher levels of reimbursement for every category studied, including physician payments. This suggests substitution against hospitals but the absence of competition in medical markets.

Age was significant in all of the regressions except the hospital admission rate model. Although the effects of increased age, both in lowering the probability of hospital admission and death in the hospital and in reducing the number of inpatient days, were small, the estimates in reimbursement models were not. Older beneficiaries, holding other variables constant, received less inpatient, Part B, and home care.


The most important finding here is that, for the first time in Medicare's history, beneficiaries are likely to die somewhere other than a hospital. Our results indicate that PPS has quickened the downward trend in the fraction of Medicare beneficiaries who die in the hospital, mainly through effects on the average length of stay. Although trends in the last 90 days of life indicated fewer days of inpatient care, general access to covered services and hospital admission rates did not decline. Indeed, it appears that PPS has simply led to more frequent discharge of dying patients to the home environment, with accompanying physician, DME, and home care services. These results are similar to but not as pronounced as those reported among terminally ill patients exposed to hospice programming (Greer, Mor, Morris, et al. 1986). The exception is that physician services appeared as a substitute for inpatient care in our sample, whereas Greer et al. found a complementary relationship between the two types of care among the terminally ill.

These general associations between PPS and patterns of practice in the 90 days before death represent important alterations in the way Medicare benefits are being used in the final stages of life. The observed changes are consistent with general PPS incentives to reduce the amount of inpatient care delivered to beneficiareis, and to increase use of home- or SNF-based services, particularly to allow lengths of stay to be safely shortened. Yet the structure of Medicare payments for the last 90 days of life is relatively unchanged, still dominated by hospitals, which continue to receive about 75 percent of Medicare payments. Whatever substitutions are being made seem not to be materially altering the structure of payments or the continued importance of hospital payments in cost-control policymaking. We also note that the observed changes in the use of hospitals as the site of death have implications for program monitoring. The use of hospital discharge mortality (or even risk-adjusted hospital mortality) as a metric for quality would seem less useful in view of the significant trends in location of death. The trends we note in the place of death suggest that hospital performance studies should prefer death rates based on days following admission.

Despite evidence of apparent substitution of nonhospital services, and no evidence of hampered access, we cannot conclude that the recent trends are leaving unchanged the quality of care delivered under Medicare's PPS. While the apparent patterns--of substituting posthospital care for shortened stays, and unchanged admission rates -- are comforting, and while the entire pattern of trends is probably consistent with the hopes of policymakers, we cannot conclude tha the health or satisfaction of beneficiaries are not affected by PPS.


1. A recent econometric study by Schmitz (1989) finds significant changes in relative length of stay (LOS) trends in PPS and waiver states in 1983 and 1984, suggesting relative reductions in LOS under PPS in those years.

2. The fact that trends are more pronounced in states under PPS is a very important design consideration in the statistical models presented later in the article. In these models the specification of PPS effects captures any divergence in trend between PPS and waivered states, whether due to a break in trend for the PPS states or for the waivered states. These data suggest that the predominant temporal change is in the PPS states, lending support to the view that PPS (or something else unique to PPS states) may have been the source of measured differences in trend between the two groups of states.

3. A set of important changes in the Medicare HHA benefit in 1981 led to widespread expansion of the industry and of use of services (Williams, Cella, and Gaumer 1984). The most important change was elimination of the requirement for prior hospitalization.

4. Since the group of persons using no services is probably dominated by instances of sudden onset of acute disease, the measure probably is an insensitive marker for changes in constraints on use. This set of persons also necessarily includes persons who may have used services but did not submit bills for reasons of the deductible or other considerations.


Garfinkel, S., G. Riley, and V. Iannacchione. "High-Cost Users of Medical Care." Health Care Financing Review 9, no. 4 (Summer 1988): 41-52.

Gaumer, G., E. Poggio, C. Collen, C. Sennett, and B. Schmitz. "Effects of State Prospective Reimbursement Programs on Patient Care." Medical Care 27, no. 7 (November 1987): 724-36.

Ginsburg, P., and D. Koretz. "Bed Availability and Hospital Utilization: Estimates of the Roemer Effect." Health Care Financing Review 5, no. 1 (Fall 1983): 87-92.

Gornick, M., and M. Hall, "Trends in Medicare Use of Post Hospital Care." Health Care Financing Review 1988 Annual Supplement (December 1988): 27-38.

Greer, D. S., V. Mor, J. N. Morris, S. Sherwood, D. Kidder, and H. Birnhaum. "An Alternative in Terminal Care: Results of the National Hospice Study." Journal of Chronic Diseases 39, no 1 (1986): 9-26.

Kidder, D., K. Merrell, and D. Dohan. Medicare Hospice Benefit Program Evaluation, Health Care Financing Grants and Contract Report. Baltimore, MD: HCFA, 1988.

Lubitz, J., and R. Prihoda. "The Use and Costs of Medicare Services in the Last 2 Years of Life." Health Care Financing Review 5, no. 3 (Spring 1984): 117-31.

McCall, N. "Utilization and Costs of Medicare Services by Beneficiaries in Their Last Year of Life." Medical Care 22, no. 4 (April 1984): 329-42.

Prospective Payment Assessment Commission. Technical Appendices to the Report and Recommendations to the Secretary. Washington, DC: DHHS, March 1988.

Riley, G., J. Lubitz, R. Prihoda, and E. Rabey. "The Use and Costs of Medicare Services by Cause of Death." Inquiry 24, no. 3 (Fall 1987): 233-44.

Riley, G., J. Lubitz, R. Prihoda, and M. Stevenson. "Changes in the Distribution of Medicare Expenditures among Aged Enrollees, 1969-1982." Health Care Financing Review 7, no. 3 (Spring 1986): 53-63.

Schmitz, R. Effects of PPS on Per Capita Medicare Utilization, Cambridge, MA: Abt Associates Inc., February 1989.

Williams, J., M. Cella, and G. Gaumer. Home Health Services: An Industry in Transition. Cambridge, MA: Abt Associates Inc., February 1984.
COPYRIGHT 1992 Health Research and Educational Trust
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Gaumer, Gary L.; Stavins, Joanna
Publication:Health Services Research
Date:Feb 1, 1992
Previous Article:Supply and demand factors in the determination of Medicare expenditures.
Next Article:Predicting length of stay for patients with psychoses.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters