Levinsky NG, Yu W, Ash A, Moskowitz M, Gazelle G, Saynina O, Emanuel EJ. Influence of Age on Medicare Expenditures and Medical Care in the Last Year of Life. JAMA. 2001;286(11):1349-1355. doi:10.1001/jama.286.11.1349
Author Affiliations: Health Care Research Unit, Section of General Internal Medicine (Drs Ash and Moskowitz), Department of Medicine (Drs Levinsky, Ash, and Moskowitz), Boston University School of Medicine, Boston, Mass; Health Economics Resource Center of Health Services Research and Development Services and Center for Cooperative Studies in Health Services, US Department of Veterans Affairs, Menlo Park, Calif (Dr Yu); Palliative and Supportive Medicine Program, Harvard Vanguard Medical Associates, Boston, Mass (Dr Gazelle); National Bureau of Economic Research, Palo Alto, Calif (Ms Saynina); and Department of Clinical Bioethics, National Institutes of Health, Bethesda, Md (Dr Emanuel).
Context Expenditures for Medicare beneficiaries in the last year of life decrease
with increasing age. The cause of this phenomenon is uncertain.
Objectives To examine this pattern in detail and evaluate whether decreases in
aggressiveness of medical care explain the phenomenon.
Design, Setting, and Patients Analysis of sample Medicare data for beneficiaries aged 65 years or
older from Massachusetts (n = 34 131) and California (n = 19 064)
who died in 1996.
Main Outcome Measure Medical expenditures during the last year of life, analyzed by age group,
sex, race, place and cause of death, comorbidity, and use of hospital services.
Results For Massachusetts and California, respectively, Medicare expenditures
per beneficiary were $35 300 and $27 800 among those aged 65 through
74 years vs $22 000 and $21 600 for those aged 85 years or older.
The pattern of decreasing Medicare expenditures with age is pervasive, persisting
throughout the last year of life in both states for both sexes, for black
and white beneficiaries, for persons with varying levels of comorbidity, and
for those receiving hospice vs conventional care, regardless of cause and
site of death. The aggressiveness of medical care in both Massachusetts and
California also decreased with age, as judged by less frequent hospital and
intensive care unit admissions and by markedly decreasing use of cardiac catheterization,
dialysis, ventilators, and pulmonary artery monitors, regardless of cause
of death. Decrease in the cost of hospital services accounts for approximately
80% of the decrease in Medicare expenditures with age in both states.
Conclusions Medicare expenditures in the last year of life decrease with age, especially
for those aged 85 years or older. This is in large part because the aggressiveness
of medical care in the last year of life decreases with increasing age.
There is great interest in the cost of medical care in the last year
of life. Expenditures for Medicare beneficiaries who die each year are approximately
5 times as high per person as for survivors.1
However, Medicare expenditures for Medicare beneficiaries in the last year
of life decrease as age increases; this pattern has been found in studies
of data from 1976,1 1978,2
1988,1 and 1992.3
To understand this phenomenon, we examined the pattern of expenditures with
age in more detail, stratifying 1996 Medicare data by sex, race, comorbidity,
use of hospice care, place and cause of death, and type of health care services
used. In addition, we have examined changes in the aggressiveness of medical
care with age, as judged by the frequency of admissions to a hospital and
to care in an intensive care unit (ICU), and by the use of aggressive interventions
such as ventilators, pulmonary artery monitors, cardiac catheterization, and
We used data from 2 states, Massachusetts and California, rather than
national data, because this made it feasible to obtain information about place
and cause of death, key variables for our analysis. In both states, we selected
Medicare beneficiaries who died in 1996, were enrolled in the Medicare program
throughout their last year of life, and were not enrolled in Medicare's End
Stage Renal Disease program. To obtain an adequate sample size for each state,
the study sample consisted of all decedents in Massachusetts and a randomly
selected 20% of decedents from the much larger state of California.
We used the denominator file from the Health Care Financing Administration
to merge with each state's 1996 death certificate files. In Massachusetts,
42 452 decedents met the above conditions. In merging with the Massachusetts
death certificate file, we used social security number, date of birth, date
of death, and sex. A match was accepted if either of the following 2 conditions
was met: (1) perfect match in social security number and either sex or both
date of birth and date of death; or (2) a match on 7 of the 9 positions in
the social security number and a perfect match in each of the sex, date of
birth, and date of death categories. This left 39 447 people (93%). To
fully characterize a person's health care use through the Medicare files,
both Part A (hospital insurance) and Part B (supplementary medical insurance)
entitlement are required. Thus, we retained only the 37 933 beneficiaries
who were continuously entitled to both kinds of insurance throughout their
last 12 months of life.
In California, we used the same method to merge the Medicare denominator
file to the death certificate file. The 20% sample contained 33 684 decedents.
Among the 33 684 people, 96% were linked to the death certificate file.
After merging with the death certificate file and excluding people who did
not meet the conditions listed above, we retained 27 658 decedents. Finally,
since Health Care Financing Administration files do not contain complete information
about health care use for beneficiaries enrolled in managed care plans, we
limited our study to people enrolled in the standard fee-for-service Medicare
program during each of the last 12 months of life. This resulted in a study
population of 34 131 decedents in Massachusetts and 19 064 in California.
The total insurance expenditure was calculated as the sum of Health
Care Financing Administration payments. When a beneficiary's primary health
insurance was not covered by Medicare, payments were calculated by what Medicare
would have paid. The average payment per person from other insurance plans
accounts for only 0.15% of the total expenditure. Out-of-pocket expenses and
Medigap copayments and deductibles are not included. Expenditure for each
person was calculated from the following files: Medicare Provider Analysis
and Review, including acute hospitalizations, long-term hospitalizations,
and skilled nursing home care, hospital outpatient, part B physician/supplier,
home health care, and hospice.
We analyzed expenditures during the last year of life by age, sex, race,
and place and cause of death. We used the National Center for Health Statistics4 classification to group decedents by cause of death.
Information from the death certificate file was used to group place of death.
We examined differences in expenditures between people who used any hospice
and those who did not. We also examined the influence of comorbidity on expenditures,
using the Charlson score5 to estimate comorbidity.
Persons were grouped into 3 levels of comorbidity: Charlson scores 0 to 2;
3 to 4; and 5 or more. In addition, we determined expenditures for each type
of health care service, such as for hospitals or physicians. Hospital expenditures
are for acute care hospitalizations only. Physician expenditures include insurance
payments for all types of physician services covered by the Medicare Part
B insurance plan.
We estimated the aggressiveness of medical care in the last year of
life from 2 types of indices, hospitalization and selected aggressive services.
Using the Medicare Provider Analysis and Review records, we identified admissions
that included care in an ICU, which we defined as including coronary care
units but excluding psychiatric ICUs. We also examined use patterns during
the last year of life for ventilators and pulmonary artery monitors, usually
used in an ICU; and for cardiac catheterization and dialysis, which are not
limited to ICUs. These treatments were identified from the surgical procedure
codes (up to 10) reported in the Medicare Provider Analysis and Review file.
All statistical analyses used SAS statistical software, version 6.11
(SAS Institute Inc, Cary, NC). Differences in expenditures (costs) or other
outcomes across the 3 age categories (65-74, 75-84 and ≥85 years) were
tested using analyses of variance. The Duncan test was used for pairwise differences
between age groups. The coefficient of variation (the mean divided by the
SD), associated with costs in these data is typically less than 1.3. Thus,
when comparing costs for 2 groups with at least 500 persons each, differences
of just under 10% are statistically significant at the P = .05 level. In fact, most comparisons of costs across age groups
involve several thousand people in each group and show differences larger
than 10%. Thus, virtually all pairwise comparisons are statistically significant,
even at the P<.001 level. For simplicity of presentation,
given the many tests performed, we note in the tables only the absence of
a statistically significant difference between the oldest and youngest subgroups
at the P = .05 level.
In addition, we used the SAS GLM procedure (SAS Institute Inc) to perform
a global test for a decrease in cost with age in California and Massachusetts.
In each state, we regressed cost on sex, race (black and white vs other),
cause of death (heart disease, stroke, cancer, chronic obstructive pulmonary
disease, pneumonia vs other), Charlson comorbidity score, and age. A trend
of decreasing cost with age is supported when the coefficient of age in the
regression is both statistically significant and negative. Suppose, for example,
that the coefficient of age is −$500. Then, among people whose sex,
race, cause of death, and comorbidity burden was the same, we would expect
80 year olds to cost $5000 less than those who are aged 70 years.
As shown in Table 1, Medicare
expenditures during the last year of life decrease with age. For all beneficiaries,
the decrease between the ages of 75 to 84 years and 85 years or older is about
twice as great as that between the ages of 65 through 74 years and 75 through
84 years. The decrease with age is noted in both men and women. The pattern
of decreasing expenditures with age was of approximately similar magnitude
in both whites and blacks. (Expenditures in each age category were higher
for blacks than for whites. We are studying this observation in a database
with a larger number of blacks.) Medicare expenditures decreased with age
both in persons who used hospice services and those who did not. Medicare
expenditures increased with increasing levels of comorbidity, as measured
by the Charlson score.5 At each level of comorbidity,
expenditures decreased with age.
We performed a global test (see Methods) to evaluate the regression
of expenditures (cost) on age, which held sex, race, cause of death, and comorbidity
score constant. Each additional year of age was associated with a $413 decrease
in cost in Massachusetts and a $408 decrease in California. Each of these
effects was significant at the P<.001 level.
The age trends were similar in the 2 states and persisted throughout
the year prior to death. Figure 1
shows the data by month from Massachusetts; data from California were similar.
The decrease with age was noted in every 30-day period in both states.
As shown in Table 2, the
pattern of decreasing expenditure during the last year of life with age was
noted for all causes of death. The trends of the data by age and diagnosis
were quite comparable in Massachusetts and California. The data also were
analyzed to determine whether the site of death influenced the pattern of
expenditures (Table 3). In both
states, expenditures in the last year of life were highest for those who died
in an inpatient setting, lower for those who died in a nursing home, and lowest
for those who died in a residence. The pattern of decreasing expenditures
with increasing age was comparable for persons dying in each of these sites.
As shown in Table 4, expenditures
for hospital inpatient care accounted for more than half of total expenditures.
Expenditures for hospital care decreased by about half between the ages of
65 through 74 years and 85 years or older. The decrease in hospital expenditures
between the ages of 65 through 74 years and 85 years or older accounted for
most of the overall age-related decrease in expenditures: 79.6% in Massachusetts
and 81.9% in California. Expenditures for outpatient care and physician services
also decreased with age. On the contrary, expenditures for care in skilled
nursing facilities and to a lesser extent for home health services increased
with age. The trends in the 2 states were comparable.
As shown in Table 5, admissions
in the last year of life, which were about 10% lower in California than in
Massachusetts, decreased progressively among the 3 age groups in both states.
Admissions were about 30% lower in those aged 85 years or older than in those
aged 65 through 74 years. The number of admissions to ICUs decreased with
age. It was more than 50% lower in those aged 85 years or older than in those
aged 65 through 74 years. The percentage of admissions that included some
care in an ICU also decreased with age, declining by about 40% in Massachusetts
and 25% in California between the youngest and oldest age groups. The frequency
of use of the other indices of aggressive services also decreased with age.
The decrease was especially striking for the oldest group (aged ≥85 years).
For that group, the use of ventilators and pulmonary artery monitors was reduced
by about two thirds. Cardiac catheterization was 85% less frequent. Dialysis
(outside the End Stage Renal Disease program, which we did not study) was
88% less frequent in Massachusetts and 75% less frequent in California. The
decrease in use of the aggressive procedures was indicated by 2 types of data.
There was a decrease in the number of all procedures per 1000 persons with
age. There also was a decrease with age in the percentage of admissions that
included some care in an ICU and those in which cardiac catheterization or
dialysis were used. The percentage of ICU admissions in which a ventilator
or pulmonary artery monitor was used also decreased with age.
We also evaluated the pattern of use of hospital services separately
for each of the 5 most frequent causes of death, which account for 75% of
deaths in Massachusetts and 80% in California (Table 6). The pattern of decreasing hospital and ICU admissions,
especially for the oldest groups, was found for each major cause of death.
The pattern of decreasing use of aggressive services with age also was generally
noted for all causes except cancer. Although the use of monitors, catheterization,
and dialysis decreased numerically between the youngest and the oldest groups
in cancer patients, the differences were not statistically significant.
We found that the pattern of decreasing expenditures with increasing
age for medical care of the elderly in their last year of life, especially
for the "oldest old" (aged 85 years or older), is pervasive. It is present
for both sexes, for both black and white beneficiaries, for those in hospice
as well as those in conventional care, regardless of the degree of comorbidity
and the cause or the place of death. The decrease in expenditures with age
in the last year of life is in large part the result of less aggressive care
with increasing age. Several observations buttress this conclusion. First,
80% of the decrease in total expenditures with age is accounted for by the
decrease in expenditures for hospital services. Second, the 50% decrease in
cost for hospitalization of beneficiaries in their last year of life is much
greater than the 30% decrease in the number of hospitalizations with increasing
age, suggesting reduced intensity of care of the older groups during hospitalization.
Finally, the various indices of aggressiveness that we evaluated confirm this
conclusion. The percentage of hospital admissions that included care in an
ICU decreased with age, as did the frequency of use of ventilators and pulmonary
artery monitors, even in persons admitted to an ICU. The use of cardiac catheterization
and dialysis (in persons not in the End Stage Renal Disease program) also
decreased with age, especially for the group aged 85 years or older. Age appears
to be a key determinant in decisions about medical care for older persons.
We examined the influence of the cause of death, which presumably reflects
the major illness during the last year of life, on the pattern of age-related
expenditures and on the use of the hospital services that served as indices
of aggressive care. Expenditures decreased with increasing age for each of
the most frequent causes. Riley et al,6 using
1979 data, made comparable observations. Thus, the pattern of decreasing expenditures
with increasing age cannot be explained by differential costs for different
causes of death. Decreasing use of aggressive care with age also was noted
for each of the most frequent causes of death. It appears that both the decreased
frequency of hospital admissions and the reduced use of aggressive therapy
account for most of the decrease in expenditures with age.
These observations are compatible with other reports that indicate decreased
intensity of care for older Medicare beneficiaries in their last year of life,
especially those in the oldest group. Using 1992 data from a national Medicare
population, Levinsky et al3 found that major
surgical procedures decreased with age; Yu et al7
found decreasing use of the ICU. Also consistent with this conclusion are
some observations from the SUPPORT study8- 10
(Study to Understand Prognosis and Preferences for Outcomes and Risks of Treatments),
which evaluated the care of seriously ill patients at 5 medical centers. The
age range of the patients was wider than in the present study; the median
age was 65 years. The SUPPORT investigators found that older age was associated
with higher rates of decisions to withhold aggressive care, after adjusting
for prognosis, patient preferences, severity of illness, and prior functional
status. Johnson and Kramer11 studied physicians'
responses to clinical scenarios involving life-threatening illness. At all
levels of probability of survival, a significant proportion of physicians
favored treating a younger patient more aggressively than an older patient
with the same likelihood of survival. From these various studies, it appears
that older patients at the end of their natural lifespan are treated much
less aggressively than younger patients. This is contrary to the opinions
of some policy analysts and ethicists12,13
who suggest that such persons are subjected to unreasonably aggressive and
expensive medical care.
There are several limitations to the generalizability of our results.
Our data are derived from Massachusetts and California. Although data from
the 2 states are comparable, it is possible they do not accurately reflect
national data. Against the likelihood of major differences, however, is the
fact that trends of expenditures and hospital and ICU admissions with age
are comparable with those reported in studies with national Medicare data.3,7 Our data also do not include persons
in managed care programs, since Medicare files do not contain complete information
about health care use for beneficiaries enrolled in a managed care program.
Medicare files also exclude individuals who received their medical care from
the Veterans Affairs medical program.
The pattern of decreasing expenditure with age may represent appropriate
clinical decisions by patients, their families, and their physicians to curtail
unreasonably aggressive care in older patients who are unlikely to benefit
from such care. The cognitive and functional status of individuals and potential
gains in functional life must be considered in deciding whether to use interventions
that may cause pain or discomfort. Moreover, life is not necessarily shortened
if aggressive care is withheld. In the SUPPORT study, the decrease in survival
with age was not due to decreased aggressiveness of care with increasing age.10,14 Alternatively, the decrease in intensity
of care with age may represent age discrimination in the use of medical care.11 The SUPPORT study concluded that many older patients
prefer aggressive therapy, a preference that often is not appreciated by their
physicians or family.10,14 Analysis
of the clinical care of individuals will be required to separate these possibilities.