Full text of Working Papers (Federal Reserve Bank of Richmond) : End-of-Life Medical Expenses, Working Paper 18-18

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

End-of-Life Medical Expenses

WP 18-18

Eric French
University College London, IFS, and
CEPR
John Bailey Jones
Federal Reserve Bank of Richmond
Elaine Kelly
Institute for Fiscal Studies and the
Health Foundation
Jeremy McCauley
University College London

End-of-Life Medical Expenses
Eric French, John Bailey Jones, Elaine Kelly, and Jeremy McCauley ∗
November 19, 2018
Working Paper No. 18-18
Abstract
In this review, we document end-of-life medical spending: its level, composition,
funding, and contribution to aggregate medical spending. We discuss how end-oflife expenses affect household behavior and economic evidence on the efficacy of
medical spending at the end of life. Finally, we document recent trends in health
and chronic disease at older ages and discuss what they might imply for end-of-life
spending and medical spending in the aggregate.

Prepared for the 9th Edition of the Handbook of Aging and the Social Sciences. Corresponding
author. Eric French: UCL, CEPR, and IFS, eric.french.econ@gmail.com. John Bailey Jones: Federal
Reserve Bank of Richmond. Elaine Kelly: Institute for Fiscal Studies and The Health Foundation.
Jeremy McCauley: UCL. We thank Amy Kelley and Jon Skinner for helpful comments. The views
expressed herein are those of the authors and do not necessarily reflect the views of the Federal
Reserve Bank of Richmond, the Federal Reserve System, the CEPR or the IFS.

∗

DOI: https://doi.org/10.21144/wp18-18

1

Section 1: Introduction
The high cost of end-of-life care has drawn scrutiny for some time. End-of-life care is often viewed
as unusually wasteful and costly, and it is widely believed that its reform could significantly improve
aggregate health care efficiency (e.g., Will, 1990; Gawande, 2014; Reid, 2017). At the household
level, end-of-life expenses can impose a significant financial burden and are often viewed as an
important motivation for saving at older ages (De Nardi, French, & Jones, 2016a).
In this review, we document end-of-life medical spending: its level, composition, funding, and
contribution to aggregate medical spending. We discuss how end-of-life expenses affect household
behavior. We then discuss economic evidence on the efficacy of medical spending at the end of life.
Finally, we describe recent trends in health and chronic disease at older ages and discuss what they
might imply for end-of-life spending and medical spending in the aggregate. We proceed as follows.
We begin in section 2 by documenting trends in mortality and the time and causes of death. As
mortality rates have fallen, especially at infancy, death has become increasingly concentrated at very
old ages. The causes of death have changed significantly as well; the fall in the death rates for
respiratory infections and heart diseases are particularly pronounced. The trends in health at
advanced ages are more mixed. On the one hand, the amount of time older individuals spend with
disability appears to have fallen. On the other, while trends in chronic disease are difficult to
measure and interpret, its prevalence has most likely risen.
In section 3, we discuss the international evidence on the cost of end-of-life care. We show that
end-of-life care is expensive: in the U.S., average medical spending from all payers during the last 12
months of life was $80,000 (in 2011, measured in 2014 dollars) and spending during the last three
calendar years of life was $155,000. Nonetheless, end-of-life spending comprises only a modest
fraction of aggregate expenditures, because the fraction of the population that dies in any given year
is small. The U.S. is in no way an outlier in terms of expenditures on end-of-life care. Cross-country
data show that the fraction of medical spending incurred during the last 12 months of life ranges
from 8.5% in the U.S. to 11.2% in Taiwan, while spending in the last three calendar years of life
ranges from 16.7% to 24.5%. The high level of medical spending immediately before death is part of
a pattern of elevated medical spending for several years prior to death; this suggests many end-oflife treatments are for chronic conditions.
In section 4, we discuss the financing of end-of-life care. In the U.S., most end-of-life expenses are
financed by the government through Medicare, the (almost) universal program for the elderly and
disabled. Medicare coverage for long-term care (LTC) expenses, however, is quite limited. Most LTC
expenses are paid either out of pocket or through Medicaid, which is means-tested. Countries differ
widely in how they fund end-of-life care, especially LTC. We find that cross-country spending
variation within individual categories, such as LTC or hospital care, is high. Interestingly, countries
that devote a relatively high share of their GDP to LTC are also those where the government most
generously funds that care. This may in part reflect substitution across providers, including informal
caregivers.

2

In section 5, we present evidence that the risk of incurring high out-of-pocket medical expenses at
very old ages is an important driver of savings. We discuss why privately provided LTC insurance is
not utilized more heavily. Potential reasons include the desire to leave bequests, adverse selection,
moral hazard (with respect to informal care), and means-tested social insurance programs such as
Medicaid.
In section 6, we review evidence on whether end-of-life care is unusually wasteful. The economic
evidence on the efficacy of medical spending at older ages is mixed. None of this implies, however,
that medical resources are being wasted in obviously futile attempts to extend the lives of dying
patients. The evidence instead suggests that most deaths were unexpected; a significant part of the
high spending on those who die is spending on sick people who were unlucky. This, of course, does
not imply that there is no room for improvement: most notably, there is evidence that continuing
the shift from conventional to palliative care would improve patient satisfaction and potentially even
health outcomes for those nearing the ends of their lives.
In section 7, we discuss the implications for medical spending on the ongoing demographic
transition. The effects of increased longevity on aggregate medical spending reflects two offsetting
forces: (i) at any point in time, older people have worse health, increasing medical spending; and
(ii) health at any given age has generally improved over time, reducing medical spending. Some
analysts have argued that end-of-life spending is central to this. Because time to death is a better
predictor of medical spending than absolute age, it is proximity to death, not age itself, that drives
medical spending in old age. This suggests that increased longevity is a “red herring,” with little
effect on aggregate expenditure. However, even if the proximity to death is the main determinant
of medical spending, increased longevity may cause the relationship between time to death and
medical spending to change. We review the evidence on this and find that the relationship between
time to death and medical spending is in fact evolving, due in part to the rise of treatable but
expensive chronic conditions in old age.
We conclude in section 8.

Section 2: Trends in Mortality, Causes of Death and Late-in-Life Medical Spending
The last century has seen changes in health, longevity, and medical care for developed countries that
have had profound effects on death and the costs associated with death. These developments
include large reductions in infant mortality and death from infectious disease. Due in large part to
these declines, death increasingly occurs at old ages and is increasingly associated with chronic
conditions. These conditions are often quite costly.
Trends in Life Expectancy and Age of Death
Figure 2.1 shows life expectancy at birth in the U.S. by year of birth, sex, and race. In 1900, life
expectancy at birth varied by race but very little by sex. White women had the highest life
expectancy at 48.7 years, followed closely by white men at 46.6 years. Life expectancy for African
3

90
80
70
60
50
40
30
White Men
Black/African American Men

20
10

White Women
Black/African American Women
2015

2010

2005

2000

1995

1990

1985

1980

1975

1970

1965

1960

1955

1950

1945

1940

1935

1930

1925

1920

1915

1910

1905

0
1900

Life Expectancy at birth (years)

American women and men lagged behind by more than a decade, at 33.5 and 32.5 years,
respectively. Between 1900 and 1950, life expectancy increased rapidly for all groups, with blacks
and women making the most rapid gains. Life expectancy continued to increase over the second
half of the century, albeit at a much slower pace.

Figure 2.1: Life Expectancy at Birth by Race and Sex, U.S. 1900-2015
Source: Centers for Disease Control and Prevention (2018a, Table 015). Data available annually from 1980
onward. Earlier data points identified by markers.

Life expectancy age 65 (years)

22.5
20.0
17.5
15.0
12.5
10.0
7.5
5.0
2.5
0.0

White Men
Black/African American Men

White Women
Black/African American Women

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

Figure 2.2: Remaining Life Expectancy at Age 65 by Race and Sex, U.S. 1950-2015
Source: CDC (2018a, Table 015). Data available annually from 1980 onward. Earlier data points identified by
markers.

These longevity increases mask differential trends in age-specific mortality rates. In particular, most
of the longevity gains in the first half of the 20th century were the result of declines in infant and
childhood mortality. For example, the U.S. infant mortality rate fell from 10% in 1915 to 3% in 1950
(Meckel, 1990). It continued to fall in the second half of the 20th century, to under 1%, but most of
the possible gains from reductions in childhood mortality had been realized by 1950. In contrast,
most of the longevity gain post-1950 has been from reductions in mortality rates at older ages. Life
expectancy at age 65, shown in Figure 2.2, increased about five years for most groups between 1950
4

and 2015. Life expectancy has grown more quickly for white Americans than for black Americans,
although the gap has narrowed somewhat in recent years.
The cumulative effect of these mortality declines is that death has become an old-age phenomenon.
Figure 2.3 shows that in 1936, 41% of those dying were 65 or older and only 5% were 85 or older.
Over time, the distribution of death ages has shifted to the right. In 2016, 73% of those dying were
at least 65 years old and 31% were at least 85.
35%
30%
25%
20%
15%
10%
5%
0%

Under 1

1–4

5–14

15–24

25–34

1936

35–44
1976

45–54

55–64

65–74

2016

75–84

85+
Age

Figure 2.3: Cross-Sectional Distribution of the Age of Death: U.S., Selected Years
Source: CDC (2018b) and Xu, Murphy, Kochanek, Bastian & Arias (2018).

Causes of Death
Changes in the age at death are intimately related to changes in the causes of death. However,
interpreting trends in the causes of death requires some care. The ability of medical science to treat
patients has improved drastically over time but so has the accuracy with which conditions can be
diagnosed. As a result, the number of deaths attributed to some diseases could rise, as deaths are
ascribed to conditions that were previously undiagnosed, while the number of deaths attributed to
other diseases could fall, as doctors reclassify from common to more specific causes of death.
Nonetheless, the primary causes of death appear to have changed in important ways. Table 2.1 lists
the top five causes of death in the U.S. for the years 1900, 1950, and 2000. In 1900, three of the top
five causes of death were related to infectious diseases. By 1950, the role of infectious diseases had
greatly diminished, with only “certain diseases of infancy” making the list. By 2000, the top five
causes of death were all chronic diseases or accidents. The trend toward death at later ages has
thus been accompanied by a trend toward death from chronic conditions.

5

Table 2.1: Top Five Leading Causes of Death in the United States
Rank

1900

1950

2000

1
2
3

Heart Disease
Cancer
Vascular Lesions

Heart Disease
Cancer
Stroke

4

Influenza/Pneumonia
Tuberculosis
Diarrhea/Enteritis/
Ulcerative Colitis
Heart Disease

Accidents

5

Stroke

Certain Diseases of
Infancy

Chronic Respiratory
Diseases
Accidents

Source: CDC (2018c).

Figure 2.4 shows age-adjusted death rates from five major causes for the U.S. over the period 19002015; as Figure 2.5 shows, since 1920 most of the age-adjusted death rate has been attributable to
these causes. Age-adjusted death rates from heart disease increased rapidly from 1900 to 1950,
before falling rapidly thereafter. While heart disease remained the leading cause of death
throughout the second half of the 20th century, the death rate from heart disease in 2015 was
approximately a third of that in 1950. In contrast to heart disease, the stroke death rate declined
throughout almost the entire period. Death rates from influenza and pneumonia were similar to
those from stroke (epidemic years excluded) until a sharp reduction in the 1940s. The rate of deaths
attributed to cancer increased gradually until the 1990s, falling thereafter.

Heart Disease

700

Influenza and Pneumonia
Cancer

Deaths per 100,000 People

600

Stroke
500

Accidents

400
300
200
100

2015

2010

2005

2000

1995

1990

1985

1980

1975

1970

1965

1960

1955

1950

1945

1940

1935

1930

1925

1920

1915

1910

1905

1900

0

Figure 2.4: Age-adjusted Death Rates for Selected Causes of Death, U.S. 1900-2015
Source: CDC (2018a, Table 017).

6

Total
Selected Causes
All other causes

Deaths per 100,000 People

2,500
2,000
1,500
1,000
500

2015

2010

2005

2000

1995

1990

1985

1980

1975

1970

1965

1960

1955

1950

1945

1940

1935

1930

1925

1920

1915

1910

1905

1900

0

Figure 2.5: Age-adjusted Death Rates for All Causes and Selected Causes of Death, U.S. 1900-2015
Source: CDC (2018a, Tables 016 and 017).
Notes: The death rate for Selected Causes is the sum of the death rates for heart disease, influenza and
pneumonia, cancer, stroke and accidents.

Trends in Chronic Disease and Poor Health
Death has become increasingly associated with chronic disease experienced in old age. This begs the
question of whether the incidence of chronic disease, and poor health in general, have also
increased.
Recent research has helped quantify the prevalence of chronic disease and its contribution to overall
health care costs. For example, Buttorff, Ruder, & Bauman (2017) found that in 2014, 60% of the
adult noninstitutionalised population in the U.S. had at least one chronic condition, with 42% having
more than one. They also found that Americans with five or more chronic conditions accounted for
12% of the population but 41% of health care costs. By contrast, the 40% with no chronic conditions
accounted for just 10% of the costs. Stafford et al. (2018), using U.K. administrative data from 2014
to 2016, reported that 46% of adults were living with one of 36 chronic conditions, with a total of
24% living with more than one. They found that approximately half of hospital admissions,
outpatient visits, and primary care consultations were for people with two or more chronic
conditions.
There is a general consensus that the incidence of chronic disease is increasing, but it is difficult to
construct a long time series for multiple conditions. Buttorff et al. (2017) found no evidence of an
increase in the prevalence of multiple chronic conditions between 2008 and 2014, but this is a
relatively short interval.
Measuring chronic conditions is challenging because there is no single accepted definition of a
chronic condition, and thus detection of chronic conditions has potentially changed over time.
Furthermore, what is classified as chronic is likely to change over time as treatment improves. One
example is cancer, many varieties of which have been transformed from acute to chronic diseases.
7

In the U.S, five-year survival rates for the most common types of cancer have increased from 50% in
1975 to 66% in 2012. There have been particularly large improvements in breast, prostate, and
colon cancer (National Cancer Institute, 2017). In England, one-year survival rates for those
diagnosed with cancer in 1971 were 50%, the same as the predicted 10-year survival rates for those
diagnosed in 2011 (Quaresma, Coleman, & Rachet, 2015).
Improved health care does not necessarily reduce the prevalence of chronic conditions. If improved
treatment allows individuals with a chronic disease (such as cancer) to survive longer, at any given
time more people will be living with that chronic condition. Furthermore, individuals who would
previously have died from other conditions are now more likely to survive to an age where they are
more susceptible to chronic diseases. For example, the number of individuals diagnosed with
dementia across the world is predicted to more than double by 2050. However, evidence suggests
that the incidence of dementia at any given age is falling. The rise in rates is instead attributable to
an aging population and people with dementia surviving longer (Derby, Katz, Lipton, & Hall, 2017).
More frequently tracked are lifestyle risk factors that contribute to chronic disease, such as smoking,
alcohol consumption, and obesity. These behaviors are usually easier to measure and are more
easily obtainable for long periods. Smoking has been on the decline across the developed world for
the past 50 years. In 1965, 42% of U.S. adults smoked. By 2016, this had fallen to 16% (CDC, 2018d).
Unfortunately, while smoking has fallen, rates of another leading cause of chronic disease, obesity,
have been on the rise. Recent results for the U.S. suggest that obesity continues to increase among
adults, although it may have plateaued among children (Hales, Fryar, Carroll, Freedman, & Ogden,
2018). U.K. evidence (Public Health England, 2017a) also suggests obesity is plateauing in children,
with lower rates when they enter school at age 4. Rates of obesity at the time children leave
primary school (age 11) have increased, but this is largely confined to the two most deprived
quintiles of the population.
A measure related to but broader than the incidence of chronic disease is healthy life expectancy, or
the number of years that an individual can expect to be in good health. In practice, trends in healthy
life expectancy are difficult to assess, in part because there is no universally accepted definition of
good health. Chernew, Cutler, Ghosh, & Landrum (2017) defined health on the basis of disability,
with disability itself defined as impairments in Activities of Daily Living or Instrumental Activities of
Daily Living. They estimated that life expectancy at age 65 increased by 1.3 years between 1992 and
2008, but nondisabled life expectancy (the number of years that an individual can expect to be
nondisabled) increased even more, by 1.8 years. The time spent living in disability by those 65 and
over thus fell by 0.5 years. 1 By contrast, there is evidence that when health is measured as the
presence of disease, the length of life spent in bad health has increased (Crimmins & BeltránSánchez, 2011). This is consistent with findings that although the incidence of disease has increased,
disability conditional on disease has fallen (Cutler, 2005). Crimmins & Beltrán-Sánchez (2011) also
reported, however, that the portion of life spent with mobility functioning losses has increased.

1

Such a finding is consistent with the well-known “compression of morbidity” hypothesis (Fries, 1980): if
medical progress increases the expected lifespan but has relatively little effect on the maximum possible
lifespan, as longevity increases so will the fraction of life spent in good health. Perhaps not surprisingly, there
is also a counterhypothesis that morbidity should expand (Gruenberg, 1977).

8

They conjectured that improvements in technology may be reducing the rate at which these losses
translate into disability.
Yet another measure of health, more subjective but more holistic, is the individual’s self-assessment.
In the U.K., between 2000-2002 and 2012-2014 expected years in self-rated good health for men
increased from 60.6 to 63.4 years, while expected years in poor health increased from 15.4 to 16.1
(Public Health England, 2017b). The results for women were similar.

Section 3: International Evidence on the Cost of End-of-Life Care
Measurement Issues
Measurement of late-in-life medical expenditures is not straightforward. People who have died or
are seriously ill cannot respond to surveys. Even if “exit interviews” of survivors or caregivers are
used to complete the survey, households are usually aware only of the expenses they pay out of
pocket. However, there can be many different types of medical services and many different payers
for them. Aggregating across expenditures and payers often requires bringing together data from
multiple sources, and relatively few countries have high-quality administrative data linking all
sources of medical care to mortality. Most countries lack any administrative data on end-of-life care
that are accessible to researchers, and those that do usually only cover limited information, such as
hospital care. Furthermore, in most countries there are both public and private payers for health
care. Merging administrative data from both types of insurers is difficult. Moreover, in some
countries, private insurers may cover only certain sectors of the population. For example, there are
special insurers in Germany for high earners and special insurers in Japan for particular occupations.
A sample from such an insurance company is not likely to be representative.
In the U.S., these issues have been circumvented by linking survey information to administrative
data in an effort to account for all payers in a representative sample. For end-of-life expenditures,
the Medicare Current Beneficiary Survey (MCBS) and the Health and Retirement Study (HRS) are the
key datasets. However, both datasets have their own specific problems. Table 3.1 summarizes
important features of both datasets.
The MCBS captures all payers of medical expenditure at a high frequency. Administrative data on
Medicare expenditures and Medicaid recipiency are linked to individual survey data. Individual
responses and Medicare reports of care often differ: the medical spending data are constructed
using a sophisticated reconciliation process. Nonetheless, the MCBS understates medical spending
relative to the national aggregates found in the National Health Expenditure Accounts (NHEA). This
is partly because the NHEA includes expenditures on research and development, administration and
public health, whereas the MCBS attempts only to measure personal health care expenditures.
However, the MCBS also appears to understate personal health care expenditures. De Nardi et al.
(2016b) found that the MCBS overall captured 86% of all Medicare payments and 79% of all
Medicaid payments, while French, Jones, & McCauley (2017a) found that in 2011 the MCBS captured
78% of personal health care payments, inclusive of all types of care and all payment sources. Underreports of medical spending do not seem to be specific to the MCBS. Using data from nine counties,
9

French et al. (2017a) found that the micro data from almost every country understates spending
relative to the national aggregates.
Table 3.1: Comparison of the HRS and MCBS Datasets
Data Source

Longitudinal Design

Sample Population

Interview Frequency
Measurement Unit
Interview Methodology
Institutional Population

HRS

MCBS

Survey, employer, and
administrative data from
Medicare, Medicaid, and other
sources available
for merging
Full panel, new cohorts
added as they (roughly)
reach age 50
Nationally representative of
those aged 50+

Survey data reconciled with
Medicare and Medicaid
administrative data

Every two years
Household, spouses included
Mix of in-person and other
Not included in initial
samples, but households
followed into institutions

Rotating panels, each panel
lasting four years
Nationally representative of the
Medicare population (captures 98%
of those aged 65+)
Every four months
Individual
In-person
Included (by proxy)

Source: French et al. (2017b).

Relative to the MCBS, the HRS interviews respondents much less frequently but over a much longer
period of time; once a household enters the survey it is tracked until all its members die. Members
are followed into nursing homes, and upon the death of a member, his or her survivors are
interviewed. The HRS has long contained information on out-of-pocket expenditures, and recently it
has been linked to both Medicare and Medicaid data. However, other payers, such as private
insurance, are excluded. A strength of the HRS is its very comprehensive set of survey questions,
including information on care provided by other family members. In combination with asset
information, these data provide a broad measure of the cost of end-of-life care, including informal
care, and how such costs impact wealth and overall well-being.
Given the differences in survey design, there are multiple dimensions along which the MCBS might
be better or worse than the HRS for measurement of medical spending. However, French et al.
(2017b) showed that for out-of-pocket expenses and Medicaid recipiency, the two surveys line up
well. One limitation of the MCBS is that it is representative only of Medicare beneficiaries, who
include Disability Insurance beneficiaries and virtually all of the age 65+ population and is thus
representative of deaths only within that population. Fortunately, 73% of all deaths are among
those 65+. The HRS covers a somewhat broader age range, with younger cohorts entering the
survey around age 50.

10

Estimating End-of-Life Medical Spending
How researchers measure medical spending in the last 12 months of life depends on the data
available. When data are available at an extremely high frequency, the most common approach is to
measure spending starting from the date of death and sum backward for 12 months (e.g., Hogan et
al., 2001). Einav, Finkelstein, Mullainathan, & Obermeyer (2018) have referred to this method as
“backfilling.”
However, in many large datasets, medical spending is available only at an annual frequency. For
decedents, this means that total medical spending in the last calendar year of life, which is spending
between January 1 and the date of death, is all that can be observed. Any comparison of the medical
spending of decedents with that of survivors will suffer from the problem that while all survivors had
12 months of spending, spending data for decedents mixes together those who died in January (and
so had only one month of spending in the “year of death”) and those who died in December (and so
had 12 months of spending), along with those dying in other months.
A commonly used way to work around this problem is that of Hoover, Crystal, Jumar, Sambamoorthi,
& Cantor (2002), who estimated the following regression using MCBS data:
(1)

𝐸𝐸𝑖𝑖 = 𝛽𝛽0 + 𝛽𝛽1 �𝑚𝑚𝑖𝑖 + 𝛽𝛽2 𝑚𝑚𝑖𝑖 + 𝛽𝛽3 𝑚𝑚𝑖𝑖2 + 𝜖𝜖𝑖𝑖

where 𝐸𝐸𝑖𝑖 is total medical spending in the last calendar year of life for individual i, 𝑚𝑚𝑖𝑖 is individual i’s
exact month of death (e.g., 𝑚𝑚𝑖𝑖 = 1 if the month of death is January, and so on), and 𝜖𝜖𝑖𝑖 is a zeromean residual. Average medical spending for the last 12 months of life can then be predicted by
estimating the coefficients for equation (1) and evaluating the estimated equation at 𝑚𝑚𝑖𝑖 = 12.

Estimates across Countries

Table 3.2 displays average medical spending at the end of life across nine countries. The estimates
in this table are taken from French et al. (2017a), who utilized a collection of micro datasets. 2 (For
the U.S. they used the MCBS.) Table 3.2 contains two measures of end-of-life spending: spending in
the last 12 months of life, estimated using the Hoover et al. (2002) method; and spending over the
last three calendar years of life. In the table, odd columns display average medical spending levels,
expressed in 2014 dollars, and even columns display the percentage of aggregate spending in that
medical spending category accounted for by decedents.

2

French et al. (2017a) sought to make the spending data as comparable across countries as possible by
focusing on data from the same years, using common estimation methods, and restricting spending to the
same service categories. They also adjusted their estimates so that mean medical spending per capita in each
of the micro datasets matched the national aggregates for its source country. Methodological details can be
found in French et al. (2017a) and its associated appendix.

11

Table 3.2: Average Spending on End-of-Life Care across Countries
All medical care, including
long-term care
Spending
% of Aggregate
(1)
(2)

All medical care, excluding
long-term care
Spending
% of Aggregate
(3)
(4)

Long-term care
Spending
(5)

% of Aggregate
(6)

Hospital care
Spending
(7)

% of Aggregate
(8)

12

Final 12 months of life
Denmark
62,672
10.95
52,286
9.97
9,059
21.74
36,554
10.01
England
------18,633
14.59
France
--36,350
8.50
--24,458
15.00
Germany
52,742
10.96
46,480
10.59
4,686
14.89
29,428
21.17
Japan
--38,942
5.93
--37,869
8.21
Netherlands
63,473
10.01
36,592
7.32
14,982
22.12
20,586
8.85
Quebec
------22,868
22.73
Taiwan
20,892
11.20
18,787
10.10
1,986
23.08
12,122
15.53
United States
80,094
8.45
59,180
7.11
14,034
18.12
35,376
9.91
Last 3 calendar years of life
Denmark
128,612
22.16
102,333
19.23
26,279
54.24
68,900
18.65
England
------39,620
29.78
France
--59,534
14.10
--34,804
22.08
Germany
95,844
21.40
80,633
19.85
15,211
36.59
40,834
29.87
Japan
--66,256
10.36
--61,027
13.50
Netherlands
123,019
19.40
68,332
14.28
54,687
44.86
35,159
15.12
Quebec
------24,074
25.65
Taiwan
41,716
24.48
37,542
22.07
4,174
54.92
23,910
34.88
United States
155,398
16.70
104,222
12.77
51,176
44.92
56,351
16.27
Source: French et al. (2017a).
Notes: “Spending” is per decedent in 2014 U.S. dollars. ‘% of Aggregate’ displays the spending as a percentage of all spending in that medical spending category
(both on decedents and survivors). ‘Final 12 months of life’ displays the average medical spending in 2011 that went to those who were in their last 12 months of
life. ‘Last 3 calendar years of life’ displays the average medical spending in 2011 that went to those who were in their last three years of life. For all countries the
year of death is 2011, apart from Denmark, which uses 2012 data, and France, which uses 2013 data. Medical spending in the last three calendar years of life is
the sum of medical spending in calendar years 2009 through 2011. Hospital spending refers to both inpatient and outpatient care, apart from France, England,
and Quebec, which only have data on inpatients. Japanese data only includes hospital, dentist, and pharmaceuticals. “Long-term care” for Taiwan also includes
home help. Data from Germany exclude home help. ‘—‘ denotes data unavailable.

Column (1) of Table 3.2 shows that average medical spending in the last 12 months of life is high,
reaching $80,000 for the U.S., $63,000 for the Netherlands and Denmark, and $53,000 for Germany.
Medical spending is also high during the last three calendar years of life, reaching $155,000 for the
U.S., $123,000 for the Netherlands, $129,000 for Denmark, and $96,000 for Germany. These totals
are roughly double those incurred in the final 12 months. Thus, the spending of those who die is far
from fully concentrated right at the time of death. This suggests that the high cost of dying is due
less to last-ditch efforts to save lives than to spending on chronic conditions, which are associated
with shorter life expectancies. In a similar vein, Davis, Nallamothu, Banerjee, & Bynum (2016), who
studied spending trajectories near the end of life, found that while 49% of U.S. decedents
experienced “high persistent spending,” only 12% had “late rise spending.”

Figure 3.1: Average Medical Spending in the Last 12 Months of Life in the U.S., by Expenditure Type.
Source: De Nardi et al. (2016c), adjusted to match aggregate spending.

Figure 3.1, taken from De Nardi et al. (2016c), contains a similar message. Figure 3.1 plots cumulative
medical spending over the last 12 months of life for the U.S. 3 Although an average of almost $40,000
is spent in the three months preceding death, another $40,000 is spent in the nine months preceding
that.
The even columns of Table 3.2 show for each country the fractions of national medical spending (in
2011) devoted to people in the last 12 months and last three calendar years of their lives. Although
dying is expensive in all countries, in all countries the fraction of the population that dies in any given
year is small. Medical expenses for those close to death therefore do not necessarily account for a
large portion of aggregate medical expenditures. Each set of fractions is specific to the medical
spending category listed in the headers. Column (2) displays the spending shares for all medical care

3

Figure 3.1 is based on data from the MCBS, adjusted for its undercount of medical spending relative to the
national aggregates.

13

services. The top panel shows that medical spending in the last 12 months of life accounts for
approximately 8-11% of aggregate medical spending in most countries, with the U.S. spending the
least (8.5%) and Taiwan the most (11.2%) in percentage terms. There is no strong link between this
percentage and the type of health care system. 4 The bottom panel shows that total medical spending
in the last three calendar years of life is approximately twice as much as spending in the last 12
months, with the spending fractions ranging from 16.7% in the U.S. to 24.5% in Taiwan.
While the finding that end-of-life spending is a modest fraction of aggregate medical spending may be
at odds with popular wisdom, it comports with earlier studies. Emanuel & Emanuel (1994) calculated
that 10-12% of total U.S. medical spending was for end-of-life care. Aldridge & Kelley (2015) found a
slightly higher value, 13%, but relied heavily on imputations. One possible source of misunderstanding
for the U.S. is the well-known finding by Lubitz & Riley (1993), reaffirmed by Riley & Lubitz (2010), that
about 25% of Medicare spending on the 65+ population is for care in the last year of life. However,
this calculation overstates the share of aggregate medical spending that goes to those in the last year
of life. Although most deaths are among those 65 and older, most medical spending goes to those 64
and younger. The share of 8.5% reported in Table 3.2 is based on end-of-life and aggregate spending
for all ages.
What is the Money Spent On?
The remaining columns of Table 3.2 display spending and spending shares for different medical service
categories. Column (6) shows that, relative to other medical services, a much greater proportion of
LTC expenditures are incurred near the end of life. The share of LTC spending incurred in the last 12
months of life ranges from 14.9% in Germany to 23.1% in Taiwan. The share incurred in the last three
calendar years of life ranges from 36.6% in Germany to 54.9% in Taiwan.
Columns (7) and (8) show spending and spending shares for hospital services, a spending category
measured in all nine countries. In the countries where every medical spending category is accounted
for, hospital care is the largest end-of-life expense. Hospital spending is not always more concentrated
at the end of life than medical spending in general, and, Germany excepted, hospital spending is less
concentrated than spending on LTC. The share of hospital spending accounted for by those in the last
12 months of life ranges from 8.2% in Japan to 22.7% in Quebec, and the share for those in the last
three calendar years of life varies from 13.5% in Japan to 34.9% in Taiwan. 5 Comparing columns (2)
and (8) shows that among countries with complete data, the variation in hospital spending shares is
larger than that of the overall spending shares. For example, in the last three years of life, hospital
shares for this group range from 15.2% to 34.9%, while the overall shares range between 16.7% and
24.9%. This suggests that some of the variation in the hospital shares is due to substitution between
providers and/or services in the delivery of care.
Figure 3.1 sorts U.S. spending over the last 12 months of life into medical service categories.
Inpatient hospital spending comprises the bulk of spending in the last two months of life. Other
4

French & Kelly (2016) briefly described each country’s health care system.
These results are in line with Bekelman et al. (2016), who compared patients older than 65 who died with
cancer across seven developed countries. They found that end-of-life care was more hospital centric in
Belgium, Canada, England, Germany, and Norway than in the Netherlands or the United States.

5

14

forms of care, such as LTC (which includes nursing home care) are more important in the months
further from death.

Section 4: Funding for End-of-Life Care
Although end-of-life spending may comprise a relatively modest fraction of spending in the aggregate,
at the household level it is hardly trivial. The averages reported in Table 3.2, high in their own right,
mask the possibility of significantly higher expenditures. In this section, we describe how end-of-life
care is funded, the first step in assessing the financial risk it poses to older households.
A country-by-country account of how end-of-life care is funded is beyond the scope of this review;
French & Kelly (2016) provided summaries for the countries appearing in Table 3.2. Below, we provide
a detailed description for the U.S., followed by a cross-country comparison of LTC, the service that
seems to have the most cross-country variation in funding.
Funding in the U.S.
Figure 4.1 plots cumulative medical spending over the last year of life for the U.S., using the same data
and methodology as Figure 3.1, but decomposing spending by payer rather than service. Of the
$80,000 incurred over the final year of life, 66% is covered by Medicare, the public health insurance
available to almost everyone 65 or older. 6 9% is covered by Medicaid, public health insurance that is
means-tested, and 2% by other government programmes. Many people have private coverage, and
thus 8% of end-of-life costs are paid by private insurers. While end-of-life care is reasonably wellinsured, there is still a nontrivial amount that must be paid out-of-pocket. Out-of-pocket expenses in
the last year of life are $9,530, or 12% of the total. 7 Additionally, even after other payers have
contributed, expenses can be so high that households are unable to cover them: uncollected liabilities
are $2,060, or 3% of the total. Nonetheless, end-of-life expenses are on average better insured than
medical spending for the elderly in general. The MCBS data behind Figure 4.1 shows that 19% of the
medical spending by the over-65 population is paid out of pocket, relative to 12% for those in the last
year of life.
Figure 3.1 shows that most end-of-life spending is for either hospital care or LTC. The largest
expenditure item is hospital care. Hospital care is well-insured for those 65 and older, as Medicare
covers most of their hospital costs, and many have private Medigap policies that pay for the
remainder. Only 1% of hospital care is paid for out of pocket (De Nardi et al., 2016c).

6

In addition to being 65 or older, an individual or their spouse must have paid Medicare taxes for at least 10
years. Individuals not meeting this requirement may be able to buy Medicare coverage. (U.S. Centers for
Medicare & Medicaid Services, 2018). 98% of the 65+ population receives Medicare (see Table 2.1).
7
French, Baker, Doctor, De Nardi, & Jones (2006) and Marshall, McGarry & Skinner (2011) found even higher
out-of-pocket expenses in the last year of life.

15

Figure 4.1: Average Medical Spending in the Last 12 Months of Life in the U.S., by Payer
Source: De Nardi et al. (2016c), adjusted to match aggregate spending.

Medicare coverage of LTC and home help costs is far less comprehensive. Medicare pays only for
skilled nursing care such as rehabilitative services, but most LTC is unskilled custodial care.
Furthermore, Medicare pays for at most 100 days in a nursing home (U.S. Centers for Medicare &
Medicaid Services, 2018). While Medicare covers nearly 70% of the hospital costs of those 70 and
older, it covers less than 25% of their nursing home costs. Furthermore, few individuals purchase LTC
insurance; private insurance covers less than 8% of nursing home costs. As a result, a significant
portion of LTC spending is uninsured: around 28% of LTC expenditures are paid for out of pocket
(De Nardi et al., 2016c). 8 Given its high cost, its concentration near the end of life (see Table 3.2), and
its high out-of-pocket share, LTC is an important driver of out-of-pocket spending near the end of life.
The largest LTC payer, covering almost 30%, is Medicaid. While the Medicaid rules are complicated
and vary from state to state, people in nursing homes typically qualify through one of two channels
(De Nardi et al., 2012). They are either “categorically needy,” because their income and wealth are
low in an absolute sense; or they are “medically needy,” because their medical expenses have
exhausted their financial resources. The latter provision extends Medicaid beyond the lifetime poor
but requires households to spend down their wealth before receiving benefits. An open question is
the extent to which wealthier households avoid spend-down through trusts, transfers, or other
financial devices. While the data show that people with higher wealth are much less likely to receive
Medicaid (Borella et al., 2018) and take longer to spend down their wealth (Wiener, Anderson,
Khatutsky, Kaganova, & O’Keeffe, 2013), they also show that at very old ages a significant fraction of
high-income individuals are on Medicaid (De Nardi et al., 2016b). 9
See Table 1. Values are based on the 2010 National Health Expenditure Accounts.
There are also a number of direct analyses of spend-down evasion. While the ability of Medicaid to recover
its costs from its beneficiaries’ estates appears quite limited (see the discussion in De Nardi et al., 2012), the
8

9

16

Differences in funding of the types of health care lead to differences in the amount of insurance
provided to those with different health conditions. Health conditions requiring LTC are relatively
poorly insured. Kelley, McGarry, Gorges, & Skinner (2015) used the HRS to estimate medical costs
over the last five years of decedents’ lives. As shown in Table 4.1, they found that the total average
expenditures per decedent for dementia ($218,000) were significantly greater than for those who died
of heart disease ($153,000), cancer ($144,000), or other causes ($164,000).
Table 4.1: Expenditures over the Last Five Years of Life, by Disease and Payer
Average
Medical
Medicare
Expenditures* & Medicaid
(1)
(2)

Medicare
(3)

Out-ofPocket
(4)

Imputed
Informal
Care
(5)

Social
Cost
(1 + 5)
(6)

Dementia

$218,288

$130,295

$92,476

$65,826

$88,830

307,117

Cancer

143,537

109,636

108,330

30,834

41,974

185,512

Heart Disease

152,877

103,266

97,769

37,763

34,510

187,387

Other

164,022

117,495

111,046

38,596

47,065

211,087

Source: Kelley et al. (2015, Table 1).
Notes: Values in Kelley et al. (2015) converted from 2010 to 2014 dollars using the PCE. The sample
includes 1,702 subjects over age 70 in the HRS who died between 2005 and 2010. *Average medical
expenditures include Medicare, Medicaid, out-of-pocket expenses, and imputed third party payments
for nursing home expenditures. Average medical expenditures exclude imputed informal care costs.

While Medicare expenditures were similar across illnesses, average out-of-pocket spending for
dementia patients ($66,000) was 81% higher than that of non-dementia patients, consistent with
Medicare’s limited coverage of LTC services. Not only was absolute out-of-pocket spending
significantly higher within the dementia group, but Kelley et al. (2015) reported that out-of-pocket
spending as a proportion of total household wealth five years prior to death was also substantially
higher, as those who die from dementia tend to have less wealth than those who die of other diseases.
The gap in the out-of-pocket burden between dementia and non-dementia decedents was especially
pronounced in lower education and minority groups.
In addition to out-of-pocket expenditures, informal care provided near the end of life can impose great
strains. Informal care is often ignored in end-of-life expenditure analyses. Imputing the implicit cost
of informal caregiving, 10 Kelley et al. (2015) found that the average informal care burden for dementia
decedents was more than double that of non-dementia decedents.

use of trusts to qualify for Medicaid appears limited as well (Taylor, Sloan, & Norton, 1999). The degree to
which households transfer assets to qualify for Medicaid, discussed below, also appears to be modest.
10
HRS respondents report the total hours of informal care provided in the month prior to each interview.
Kelley et al. (2015) converted these reports into five-year totals, which they then multiplied by the statespecific average cost of home help services.

17

Differences in Long-Term Care Systems
While many medical services impose significant financial burdens on U.S. households, LTC is arguably
the most significant. It turns out that LTC is a significant financial risk for households in many other
countries as well. As Brown & Finkelstein (2011) noted, although most OECD countries other than
the U.S. provide universal insurance for acute care, many fail to provide similar insurance for LTC. 11
There is also considerable cross-country variation in LTC expenditures. Figure 4.2 shows that total
spending on LTC equals almost 3% of GDP in some countries, such as the Netherlands and Sweden.
Other countries, such as Greece, Portugal, and Poland, devote less than .4% of GDP to LTC. The LTC
share for the U.S. is toward the bottom of the range at .9% in 2016, in contrast to the overall U.S.
health expenditure share, which is much higher than in other countries. Variation in funding
arrangements may be driving the dispersion in LTC shares. 12 For example, Scandinavian countries
and the Netherlands, which provide universal LTC, spend a very high share of GDP on LTC. In
contrast, the U.K., Canada, and the U.S., who have means-tested public programs for LTC, spend a
more modest share of their GDP on LTC.

Total LTC Expenditure
3.5
3

Public LTC Expenditure

2.5
2
1.5
1
0.5
0

Figure 4.2: Long-Term Care Expenditures in the OECD, as Percentages of GDP, 2016
Source: OECD (2018). This figure is an updated version of Figure 1 in Brown & Finkelstein (2011).
Notes: Figure shows total expenditures and public LTC expenditures as percentages of GDP for various OECD
countries. Public LTC expenditure data for the U.S. are not available for 2016; this spending is inferred by
assuming that the public share of total LTC expenditures in 2016 was the same as in 2013.

Some of the differences in spending likely reflect substitution across providers, particularly to or
away from informal care. Rodrigues, Huber, & Lamura (2012, Figure 7.4) used cross-country
variation to show a negative correlation between formal care provision and informal care provided

11

Descriptions of LTC systems can be found in Dobrescu (2015) and Nakajima & Telyukova (2016).
Public financing of LTC has been shown to have an impact on use of care at end-of-life. Orlovic, Marti, &
Mossialos (2017) found that in countries where public financing of LTC is particularly strong, patients at the
end-of-life were more likely to have reduced hospitalizations and a higher share of out-of-hospital deaths.

12

18

by families: for example, individuals in both Sweden and the Netherlands receive relatively little
informal family care and a relatively large amount of formal care. Barczyk & Kredler (2018) showed
that as one moves from north to south across Western Europe, public funding of LTC falls and the
use of informal care rises; the U.S.’s location on this gradient is somewhere between the central and
southern European countries. Accounting for informal care only reinforces the conclusion that LTC
spending risk is not unique to the U.S. Indeed, in countries such as the U.K., there have been policy
proposals to provide greater state support for LTC (Commission on Funding of Care and Support,
2011).
It is natural to think that because LTC spending tends to be concentrated toward the end of the life
cycle, countries with relatively high LTC spending would also incur more of their medical spending
late in life. However, comparing Figure 4.2 to Table 3.2 (column 2) provides little if any support for
this conjecture. For example, the Netherlands devotes relatively little of its medical spending to
end-of-life care, despite spending a large share of its GDP on LTC. Conversely, Taiwan devotes a
relatively large share of its medical spending to end-of-life care, despite spending little on LTC.

Section 5: End-of-Life Care as a Driver of Saving and Other Financial Behavior
Having established that uninsured end-of-life spending, particularly for LTC, often imposes a
significant financial burden, we turn to the evidence on how households respond to this risk. In
addition to its direct effects, end-of-life medical spending may provide an important motive for
retirement saving (De Nardi et al., 2016a). That is, elderly households may be holding onto their
assets to cover expensive medical conditions at extremely old ages. While late-in-life medical
expenses mechanically reduce wealth, the risk of catastrophic late-in-life medical spending may lead
to saving that increases wealth.
Wealth at the End of Life
Several papers have shown that death is associated with significant declines in household wealth
and that these declines are driven at least in part by high medical spending around the time of
death. French et al. (2006) used HRS data to document changes in health care use, medical
spending, and assets around the time of death. Poterba, Venti, & Wise (2011) also showed that
wealth declines around the time of death.
De Nardi et al. (2018b) updated the results in French et al. (2006) and developed a model of lifetime
decision-making to help understand them. Figure 5.1, taken from this study, shows the medical
spending of couples and singles, respectively, around the time of death. Each panel of this figure
compares the spending of households that experience a death with the spending of households with
similar initial household composition (i.e., single man, single woman, couple), health, age, income,
and wealth that do not experience a death. Medicaid as well as out-of-pocket spending is included,
as Medicaid is not available unconditionally but is subject to means-testing.
The left panel of Figure 5.1 shows average annual medical spending for married households that lose
a spouse from six years prior to death to four years after death. The right panel shows average
19

annual spending for singles (including those widowed, divorced, and never married) in the six years
prior to their deaths. Six years prior to a household death, average out-of-pocket plus Medicaid
spending is $8,000 per year for couples. This spending rises in the years leading up to death,
reaching $19,000 in the period the death is recorded. 13 After the period of death, medical spending
returns to its original level. The right panel of Figure 5.1 shows that for singles, medical spending
rises from $6,000 six years prior to death to $16,000 in the year the death is recorded.
Annual OOP plus Medicaid Expenses

2014 dollars

20000

30000

Around Death, Initial Singles

0

10000

30000
20000
10000
0

2014 dollars

Annual OOP plus Medicaid Expenses

Around Death of a Spouse, Initial Couples

-6

-4

-2

Years

0

Death in Household

2

4

-6

-4

No Death in Household

Years
Treatment

-2

0

Control

Figure 5.1: Average Annual Out-of-pocket, Medicaid and Death Expenses around the Time of Death
Source: Reproduced from De Nardi et al. (2018b).

Assets

Assets

Around Death of a Spouse, Initial Couples

-2

Death in Household

Years

0

2

4

No Death in Household

110

120

130
-4

100

1000s 2014 dollars
-6

90

400
300
200

1000s 2014 dollars

500

Around Death, Initial Singles

-6

-4
Death in Household

Years

-2

0

No Death in Household

Figure 5.2: Median Wealth around the Time of Death
Source: Reproduced from De Nardi et al. (2018b).

Figure 5.2 shows median assets around the time of death. As with the graphs for medical spending
in Figure 5.1, the left panel is for households who begin as couples and transition to singles, and the
right panel is for households who begin as singles and then die. Similarly, Figure 5.2 shows asset
profiles both for households who experience a death, as well as for otherwise similar households
who do not. Households that experience a death have more rapid wealth declines in the years prior
to the death. These wealth declines are significant. For example, in the 10 years surrounding the

13

The HRS medical spending measure is backward-looking: respondents are asked about the medical expenses
they (or the decedent) recorded in the two years preceding the interview.

20

death of the first member of a couple, median assets fall by approximately $100,000, whereas
couples who do not experience a death experience no wealth declines.
A perhaps surprising result in Figures 5.1 and 5.2 is that the drop in assets around the time of death is
too large to be explained by measured medical expenditures, especially for couples. Possible
explanations include the misclassification of health-related expenditures as regular consumption (e.g.,
home remodeling) or inter-vivos transfers, perhaps in exchange for informal care. 14
Saving Behavior
Many elderly households draw down their wealth at a relatively slow rate. A commonly offered
explanation for this behavior is that older households are saving against the possibility of high
medical expenses near the ends of their lives (De Nardi et al., 2016a). De Nardi et al. (2009, 2010)
and Kopecky & Koreshkova (2014) showed that life-cycle models calibrated to observed patterns of
mortality and medical spending can explain a significant portion of U.S. saving during retirement.
Medical spending is not the only potential explanation of the elderly’s thrifty behavior. Households
may be saving to leave bequests or to ensure continued residence in their own homes.
Distinguishing these motives in the data is difficult. For example, even if the elderly save solely to
cover their medical expenses, many of them will leave bequests because they die earlier or face
lower medical expenses than expected. It is also likely that the motives operate concurrently
(Dynan, Skinner, & Zeldes, 2002), so that the question is one of relative strength. As De Nardi et al.
(2016a) noted, differentiating the competing explanations requires examining features of the data
beyond wealth and consumption, such as insurance take-up.
One strategy for assessing medical spending motives is to exploit cross-country variation. Nakajima
& Telyukova (2016, 2018) compared the saving behavior of older households in countries that
provide universal public LTC insurance with the behavior of households in countries that do not.
They found that saving is in fact lower in countries with universal public LTC insurance, but that most
of the effect was on financial assets, not housing. 15
The distinction between housing and nonhousing assets is an important one. Households appear to
run down their financial assets much more quickly than they run down their housing wealth.
Nakajima & Telyukova (2017) concluded that older households place a high value on residing in their
own homes. Because this utility flow ceases once individuals enter a nursing facility, home equity is

14

Kvaerner (2018) studied Norwegian households where a member has been diagnosed with cancer. He
found that after a cancer diagnosis, single households make significant inter-vivos transfers but married
households do not. (In interpreting these results, note that Norway provides universal LTC insurance.)
15
Blundell, Crawford, French, & Tetlow (2016) compared savings patterns of older household in the U.S.to
those in England, which has a similar system of funding for LTC but fully insures other health care. They found
faster asset growth in England than the U.S. The differences appeared to be largely driven by increases in
English house prices. Focusing only on financial assets, they found that Americans run down their financial
assets more slowly than the English.

21

a particularly effective asset for guarding against nursing home expenses (Davidoff, 2010). 16 A
potential alternative explanation is that, in the U.S., older households can often qualify for Medicaid
even if they own their own home, so long as they have little in the way of financial assets.
Medicaid and Insurance
In countries that fail to provide universal public LTC insurance, households may be able to insure
themselves in other ways. One way is to run down assets and qualify for means-tested public
insurance programs such as Medicaid. Medicaid is available to those with low assets (with
exemptions such as housing) and either low income or catastrophic medical expenses. (See De
Nardi et al. [2012] for more on the asset limits and other aspects of Medicaid eligibility for the
elderly.) While many individuals run down their assets paying for their nursing home care, there is
concern that others may intentionally run down their assets prior to entering a nursing home in
order to qualify for Medicaid. Most of the evidence suggests that intentional asset rundown is
modest. As noted above, elderly Americans appear reluctant to run down their assets in general.
Furthermore, the literature suggests at best modest transfers from parents to children to qualify for
Medicaid. For example, Norton (1995) concluded that elderly individuals are more likely to receive
transfers in an attempt to avoid Medicaid than to give transfers in order to receive Medicaid. While
Bassett (2007) and Baird, Hurd, & Rohwedder (2014) found that the self-assessed probability of
entering a nursing home helped to predict asset transfers, Waidmann & Liu (2006) and Bassett
(2007) found that the amounts involved were small.
Medicaid pays nursing homes less than what nursing homes typically charge, which potentially
results in lower quality. Several papers have assessed whether the LTC that is affordable with
means-tested insurance is of adequate quality. LTC, especially at nursing homes, is a combination of
medical and nonmedical goods, and the latter can differ greatly in quality, the choice between a
single and a shared room being just one dimension. Studying the nursing home industry in
Pennsylvania, Hackmann (2017) concluded that low Medicaid reimbursement rates led to lowerquality care. Ameriks et al. (2011, 2017) surveyed customers of the Vanguard financial services
company, a set of middle- to high-income individuals, about their willingness to purchase
hypothetical financial products. They concluded that the desire to avoid Medicaid-funded care is a
powerful saving motivation.
These results should not be taken to imply that Medicaid has little value. Braun, Kopecky, &
Koreshkova (2017) found that means-tested social insurance programs like Medicaid are an effective
way of insuring the elderly against late-in-life risks. De Nardi et al. (2016b) found that most elderly
singles, including wealthier ones, valued Medicaid above its actuarial cost.

16

The high value that elderly households appear to place on remaining in their own homes raises the question
of why reverse mortgages are not more heavily utilized. Nakajima and Telyukova (2017) concluded that
“[b]equest motives, uncertainty about health and expenses, and loan costs account for [the] low demand.”

22

Long-Term Care Insurance
Households also have the option to purchase medical spending insurance, particularly against LTC. If
LTC poses a significant financial risk, we might expect to see extensive use of LTC insurance products.
In practice, only about 10% of older households hold private LTC insurance (Lockwood, 2018).
Lockwood found that insurance use rose with wealth but was below 20% for all groups. Barczyk &
Kredler (2018) found that in Europe very few people purchase LTC insurance unless they are
compelled to do so. They found relatively little variation across countries: in countries with limited
public LTC insurance, households relied more heavily on informal care.
Some explanations of why households do not purchase LTC insurance involve arguments that they
can self-insure against LTC risk by saving. Savings at older ages not only provide insurance against
the risk of living long and having high medical spending, but also allow individuals to leave bequests
to their heirs. Lockwood (2018) argued that the low rate of insurance use implies that older
households are more concerned with leaving bequests than with avoiding LTC spending risk.
Another self-insurance mechanism is owner-occupied housing, which can be a good substitute for
LTC, as it is a store of wealth that can be liquidated when the individual enters a nursing home.
Davidoff (2010) described how illiquid owner-occupied housing might serve as a substitute for
formal LTC insurance. Achou (2018) found, however, that the quantitative magnitude of this effect
was small.
The reasons why households concerned with LTC risk might choose to not purchase private LTC
insurance go beyond self-insurance. Many of these explanations are based on inefficiencies in the
private market for LTC insurance.
First, there is the problem of adverse selection: if insurance prices do not fully reflect applicants’
risks, then those most likely to need LTC are the ones most likely to apply for coverage. Hendren
(2013) estimated that 23 percent of 65-year-olds have health conditions that preclude them from
purchasing LTC insurance. Those not denied coverage will still face high prices. (See Finkelstein &
McGarry, 2006, and Oster, Shoulson, Quaid, & Dorsey, 2010.)
Second, there is the problem of moral hazard: households holding private LTC insurance may switch
from informal to formal LTC (Pauly, 1990), which drives up the cost of LTC insurance. Konetzka, He,
Guo, & Nyman (2014), Mommaerts (2016), and Ko (2018) found empirical support for this
hypothesis in the HRS. The low profitability of U.S. LTC insurance providers suggests that moral
hazard problems could be severe. In spite of daily and lifetime benefit limits, LTC policies have
performed poorly. Underwriting has slowed, and many firms have left the market. It seems likely
that moral hazard is one reason why insurers underestimated their claims (Cohen, Kaur, & Darnell,
2013). It is perhaps not surprising that most new LTC insurance policies are hybrid products,
especially those that combine life and LTC insurance (Bodnar, 2016). Under these policies, LTC
benefits reduce death benefits, generating an incentive to utilize informal care. If elderly
households prefer informal care by family members to formal care, they may be unwilling to
purchase insurance products that encourage the use of formal care (Pauly, 1990). The incentives
offered by hybrid policies may better align with their wishes. In addition, the risks covered by hybrid
23

products are often negatively correlated, making them easier to insure in combination (Bodnar,
2016).
The effect of adverse selection and moral hazard, along with other supply-side imperfections, is to
make LTC insurance much less desirable. LTC insurance is expensive: Brown and Finkelstein (2011)
found that premia for LTC insurance policies are marked up substantially above expected claims,
with loads on typical policies from 13 to 66 cents on the dollar, depending on whether one accounts
for lapsed policies. These loads are much higher than the loads estimated in other private insurance
markets. Moreover, most LTC insurance policies provide only limited insurance against nursing
home risk. The typical LTC insurance contract caps both the maximum number of days covered over
the life of the policy and the maximum daily payment for a nursing home stay, a daily payment that
is often fixed in nominal terms (Fang, 2014). Even the policies that index the daily maximum
payment are typically linked to aggregate price indexes rather than actual nursing home costs,
generating substantial purchasing power risk between the time a person purchases the policy and
the time she enters a nursing home. Most available policies thus provide only modest insurance
against the risk of catastrophic LTC expenses. Brown & Finkelstein (2009) reported that
comprehensive insurance contracts exist but are not purchased. Finally, people holding LTC
insurance face the risk of unilateral price increases or insurer default (Ameriks et al., 2018).
A third potential explanation for the low take-up is the demand-side explanation that Medicaid
crowds out private LTC insurance. Medicaid is the “payer of last resort,” covering only the expenses
not covered by other insurers. This means that for many households, purchasing private LTC will
serve mostly to displace Medicaid payments. Brown & Finkelstein (2008) calculated that Medicaid
significantly reduced the return on LTC insurance for 75% of U.S. single households.
Some recent papers have considered multiple explanations simultaneously in order to determine the
key drivers of the low take-up of insurance. Braun et al. (2018), who developed a quantitative model
of the LTC insurance market, concluded that the most important reason for LTC insurance take was
applicant rejection due to adverse selection. Medicaid and administrative costs also played
important roles. Ameriks et al. (2018) focused on Medicaid and the fact that individuals receive little
satisfaction from consumption when they are sick. They also considered policy design. Exploiting
strategic survey responses, they concluded that demand would be significantly higher if LTC policies
were better tailored to households’ needs. In particular, benefits based on a person’s health
(typically measured as failures in Activities in Daily Living), rather than explicit LTC services, appeared
to be much more desirable. By allowing beneficiaries to compensate informal caregivers, such
policies would also help address issues of moral hazard.

Section 6: Evidence on the Efficacy of End-of-Life Care
In section 3, we documented that around one-tenth of all medical spending occurs in the last 12
months of life. In the U.S., one-quarter of spending by Medicare, the principal insurer of the elderly,
occurs in the last 12 months of life (Lubitz & Riley, 1993; Riley and Lubitz, 2010). These fractions,
and the high cost of dying in general, are commonly interpreted as evidence of waste. However, to
take this as evidence of waste, we must presume three things: (i) we know ahead of time who will
24

die in the next 12 months; (ii) end-of-life care does not extend the length of life; and (iii) the care
does not improve the quality of life.
Regarding the first presumption, Scitovsky (1984) was one of the first to point out that most analyses
of the cost of dying are ex post; they assess spending on people who in fact have died. To properly
uncover wasteful spending requires an ex-ante approach, that is, an assessment of spending on
people who were expected to die. In the past, most people died of diseases that killed them quickly.
Today, modern medicine can keep people alive long after being diagnosed with the diseases that
eventually kill them. Given that the timing of death is far from certain, what determines whether an
individual is “near death” (Gawande, 2014)?
Einav et al. (2018) estimated mortality risk by applying machine learning techniques to detailed
Medicare records. They found that less than 5 percent of Medicare beneficiaries who died in 2008
had, on January 1, 2008, a predicted annual mortality risk above 50 percent. In short, from (at least)
the perspective of the data available in the Medicare claims data, most deaths can be viewed as
unexpected.
Einav et al. (2018) also pointed out that most of those who die are sick, and sick people ‒ including
those who recover – use more health care than the healthy. They estimated that between 30% and
50% of the additional spending on the dead can be attributed to the higher cost of treating the sick.
It is therefore not at all obvious that spending on the ex-post dead is mostly spending on the ex-ante
“hopeless.” Making such an inference would require detailed knowledge about the impact of
specific treatments on individual survival probabilities. Kelley & Bollens-Lund (2018) have discussed
the development of a system to better predict the need for palliative care.
The second presumption, that additional health care does not extend life, is surprisingly debatable.
Although it seems uncontroversial that having access to some health care is better for health than
having no health care, it is less clear that increasing (or modestly decreasing) medical care from its
current level would significantly affect health, especially at older ages.
Most economics studies that exploited plausibly exogenous variation in health care have found little
or no health benefit from the additional health care provided by a more generous health insurance
plan (Brook et al., 1983), access to Medicare health insurance (Finkelstein & McKnight, 2008; Black,
Espín-Sánchez, French, & Litvak, 2017), higher local Medicare spending (Fisher et al., 2003), or access
to Medicaid (Finkelstein et al., 2012). However, a few studies have found that medical expenditures
have significant effects on health and/or survival: Card, Dobkin & Maestas (2009) and Doyle (2011)
are examples.
Many studies have found that the privately insured are in better health and live longer. However,
Black et al. (2017) showed that these associations are mostly driven by selection. Healthy people are
more likely have strong employment histories and thus receive insurance from a current or past
employer. They are also more likely to purchase private insurance.
To get around these problems, Khwaja (2010) estimated a structural model in which medical
expenditures both improve health and provide utility. He found that medical utilization would
25

decline by less than 20 percent over the life cycle if medical care was purely mitigative and had no
curative or preventive components. Blau & Gilleskie (2008) also estimated a structural model and
reached similar conclusions.
Most of these results suggest that a nontrivial portion of medical spending on the elderly is for “flatof-the curve” treatments. However, such evidence does not directly answer the question of
whether medical spending at the end of life is particularly ineffective. One area where end-of-life
care may be mismanaged is in the mixture of conventional and palliative care. For patients with an
advanced or terminal illness, there is evidence that those who receive palliative care can live at least
as long if not longer than those who receive conventional medical care alone (Temel et al., 2010;
Connor, Oyenson, Fitch, Spence, & Iwasaki, 2005). Temel et al. (2010) found that among patients
with stage IV lung cancer, the half randomized to be treated by a palliative care specialist while
receiving conventional oncology treatment stopped chemotherapy earlier, experienced less
suffering at the end of their lives, and lived 25% longer than those who received oncology treatment
alone. The evidence on whether a switch toward palliative care at the end of life also reduces costs
is more mixed (Teno et al., 1997; Teno, Gruneir, Schwartz, Nanda, & Wetle, 2007; Teno et al., 2013;
Krakauer, Spettell, Reisman, & Wade, 2009; Spettell et al., 2009). Cost impacts will likely vary across
both patients and the treatments considered.
The debate about whether high end-of-life spending is wasteful typically focuses on whether the
spending would prolong life. However, treatments with little direct mortality benefit may
significantly improve the quality of patients’ lives (Emanuel & Emanuel, 1994). To conclude that
end-of-life care is wasteful therefore requires a third presumption: that the treatment provided at
the end of life fails at advancing the quality, as well as the quantity, of life.
There is evidence that intensive treatment at the end of life can lead to poor quality of life for both
the patient and their caregivers (Wright et al., 2008). However, this must again be balanced against
uncertainty over when the end of life is coming. Some patients are willing to trade off current
discomfort for the hope of a cure or more time. The presumption that patients wish to extend their
lives can nonetheless mean that too little attention is given to other priorities, such as avoiding
suffering, remaining mentally aware, spending time with friends and family, and not imposing
burdens on others (Singer, Martin, & Kelner, 1999; Steinhauser et al., 2000). Treatments that
conflict with these broader priorities may be wasteful.
It bears reiterating that the alternative to prolonging life is often not the withdrawal of care
altogether (Gawande, 2014). Palliative care aims to ensure that individuals can live their remaining
lives to the fullest. This does not mean doing nothing. Palliative care packages may involve visits
from palliative care specialists, residential stays, and drugs and equipment that relieve suffering.
Patients who receive palliative care may in fact live longer than similar patients who stay within
conventional medicine (Temel et al., 2010; Connor et al., 2006).
An increased knowledge of how intensive treatment at the end of life affects patients and their
families has been accompanied by a decline in the share of patients who die in the hospital. CDC
data (CDC, 2018e) show that in 2000 around half of deaths in the United States occurred in
hospitals; by 2016, this had fallen to 36%. Over the same period, the share of deaths that occurred
26

at home increased from 23% to 30%. Deaths in hospice facilities were first recorded under a
separate category in 2003, at which time they accounted for less than 1% of deaths; by 2016,
hospice deaths accounted for almost 8% of all deaths. The growth in deaths at hospice facilities
reflects a much broader growth in the use of hospice care, as two-thirds of hospice patients in 2013
died at home. Between 2001 and 2007, the fraction of Medicare decedents accessing hospice for
three days or more rose from 19% to 30%. For decedents with a cancer diagnosis, the increase was
from 27% to 43% (National Hospice and Palliative Care Organization, 2014). In 2016, 48% of all
Medicare decedents had received at least one day of hospice care and were enrolled in hospice at
the time of death (National Hospice and Palliative Care Organization, 2018).
Uncertainties in whether curative treatment will be successful, and the value of curative and
palliative care when it is not, mean that is hard to determine whether medical spending at the end
of life is too high or too low. This applies to both individual patients and whole populations. In spite
of these uncertainties, there is a strong argument for focusing on how to improve the quality of care
for the growing elderly population, for whom care provided at the end of life can be both costly and
painful. A start would be to refrain from using therapies that do not improve quality of life, including
the use of feeding tubes in patients with dementia (Gozalo et al., 2011; Mitchell, Mor, Gozalo,
Servadio, & Teno, 2016; Mitchell et al. 2003) and “burdensome” transitions in the place of care
immediately before death (Gozalo et al., 2011).
One way to increase the probability that patients die in a place and manner of their choosing is for
them to specify their choices by drawing up an Advance Care Directive. Since the Patient
Determination Act in 1990 (Abele & Morley, 2016), U.S. patients have been able to choose whether
to have medical treatment or not, to make advance care directives, and to transfer their decisionmaking power to a friend or relative. Since 2016, Medicare has paid physicians to discuss end-of-life
issues in order to help patients draw up their directives. Between 2000 and 2010, the share of those
over 65 who died with an advance care directive in place increased from 47% to 72% (Silveira,
Wiitala, & Piette, 2014). A systematic review of advance care planning has shown it can reduce
hospitalization without increasing mortality, reduce the share of deaths that occur in hospitals, and
reduce burdensome treatments (Martin, Hayes, Gregorevic, & Lim, 2016). Overall, “advance care
planning [is] considered an essential step for achieving a ‘good death’ in which physical pain and
emotional distress are minimized, and the patient’s and family members’ treatment preferences are
respected” (Carr & Luth, 2017).
We finish this section by noting that much has been made of the incentives faced by physicians in
the U.S. The interaction between the patient and the doctor is structured more like a retail
transaction than in other countries, with more of a “the customer is always right attitude.” This may
lead physicians to err on the side of over-optimism (Baile, Lenzi, Parker, Buckman, & Cohen, 2002).
Doctors are paid for chemotherapy given and surgeries performed but not for the time taken to talk
to patients about whether further treatment is the right course of action (Gawande, 2014).
Insurance companies have been successfully sued for restricting access to treatments for the
terminally ill, even when those treatments are subsequently shown to be ineffective (Stadtmauer et
al., 2000).

27

All of this, it is claimed, has led to overtreatment of the dying in the U.S. However, it is important to
recall that the share of total medical expenditures devoted to those in the final year of life in the U.S.
is very similar to the shares for a range of other developed countries, where financial incentives for
“overtreatment” are not nearly as strong. The U.S. spends more than other countries on both the
living and those close to death. The causes of high spending at the end of life are therefore unlikely
to be specific to the U.S. and more likely to reflect universal factors, such as the range of available
treatments or the unpredictability of how patients will respond to them (Gawande, 2014).

Section 7: Demographic Transitions and Late-in-Life Medical Spending
To what extent does increased longevity lead to increased medical spending, aggregate or per
capita? Two factors come into play. The first is that older people have worse health and higher
medical expenses. Using the same MCBS sample as Figures 3.1 and 4.1, Figure 7.1 shows that
between ages 65 and 100 average annual medical spending rises by a factor of 5, from less than
$10,000 to nearly $50,000. The second factor is that over time health at any given age has generally
been improving − healthy life expectancy has generally been increasing as quickly as life expectancy
in general. With the two factors working in opposite directions, the overall trajectory of medical
spending is uncertain.

Figure 7.1: Average Medical Spending in the U.S., by Age and Expenditure Type
Source: De Nardi et al. (2016c), adjusted to match aggregate spending.

An illuminating extreme is the “red herring hypothesis” (Zweifel, Felder, & Meiers, 1999), which is
that time to death rather than absolute age determines medical spending. If the relationship
between time to death and medical spending is stable, increased longevity should have only modest
effects on aggregate expenditures. In fact, if people live longer but face the same medical expense
flows before and during their terminal illnesses, increased longevity should lead per capita medical
spending to fall, as a greater fraction of the population will be in good health.
28

If true, the red herring hypothesis implies that medical expenses rise with age only because older
individuals are more likely to die. But it is also possible that people require more expensive medical
services at older ages regardless of their proximity to death. A robust finding in the literature is that
time to death is a stronger predictor of expenditures than age; see Howden and Rice (2018) for a
recent review. However, there is also evidence that time to death is in fact a proxy for morbidity
(de Meijer, Koopmanschap, d’Uva, & van Doorslaer, 2011; Howdon & Rice, 2018) and that the cost
of dying varies with age. Yang, Norton, & Stearns (2003) found that inpatient expenditures incurred
near the end of life were higher at younger ages, while LTC expenditures rose with age. Braun et al.
(2015) found that total end-of-life costs rose with age. Reviewing the literature, Payne, Laporte,
Deber, & Coyte (2007) concluded: “On one hand, decedents’ costs rise with age for some services
and fall with age for others and, in some cases, rise to a certain age and then fall thereafter. On the
other hand, survivors’ costs generally rise with age for all services.”
An equally important consideration is that increases in longevity may be accompanied by changes in
the causes of death and thus changes in the cost of dying. Discerning these trends, however,
requires extended longitudinal data with end-of-life related information. Such data are generally
unavailable. One exception is Seshamani & Gray (2004), who estimated the rate at which hospital
expenditures rose as people approached their deaths. Comparing estimates from cross-sectional
samples from 1970, 1980 and 1990, they found that the run-up in costs became less pronounced
over time, although the differences were not statistically significant.
Measures of healthy life expectancy present a mixed message. People appear to be spending more
of their lives with chronic conditions but less with disability. Cutler (2005) argued that this reflects
increasingly “intensive medical technology.” If so, the observed reductions in disability time may not
translate into significant cost savings. On the other hand, it is suggestive, if only weakly, that our
estimate of the fraction of U.S. medical spending devoted to care in the last year of life is slightly
smaller than that estimated by Emanuel & Emanuel (1994) over 20 years ago. Riley & Lubitz (2010)
likewise concluded that the share of Medicare expenditures incurred in the last year of life has
changed very little over a 30-year period.

Section 8: Conclusion and Directions for Future Research
There is a widespread belief that end-of-life care represents large and low-hanging fruit for those
interested in health care reform. We find relatively little support for such a belief. For the countries
considered above, the fraction of aggregate medical expenditures incurred during the last 12
months of life ranges from 8.5% to 11.2%, and the fraction for the last three calendar years of life
ranges between 16.7% and 24.5%. The empirical literature suggests that a significant portion of endof-life treatment addresses chronic conditions and is unable to show that a large fraction of end-oflife spending is directed toward “hopeless” cases.
This in no way implies that there are no pressing issues. End-of-life care is expensive in absolute, if
not relative, terms. The fact that wasteful spending is hard to document does not mean that it does
not exist. The evidence suggests that a shift toward palliative care would significantly improve the
29

quality, and possibly even the quantity, of life for those with terminal diseases. Finally, in countries
without universal public LTC insurance, end-of-life care imposes significant financial risks on older
households and their informal caregivers.
We thus see a number of areas for productive future research. One is to continue to study the
efficacy and cost of end-of-life treatments. A second is to continue to study the financial
implications of end-of-life spending risk, along with proposals to mitigate it. Barczyk &Kredler (2017)
concluded that subsidizing informal care is a particularly effective strategy. More analyses of this
sort are essential. Also essential is continued research into why private LTC insurance is so lightly
subscribed. Ameriks et al. (2018) argued that basing this insurance strictly on health outcomes,
rather than the purchase of particular LTC services, would increase its demand. Finally, we have
little knowledge of how end-of-life spending, and spending at older ages in general, will evolve in the
future. Given the fiscal magnitude of public health insurance, guidance on this topic is of first-order
importance.

30

References

Abele, P., & Morley, J. (2016). Advance directives: The key to a good death? Journal of the American
Medical Directors Association, 17(4), 279-283.
Achou, B. (2018). Housing liquidity and long-term care insurance demand: A quantitative evaluation.
Retrieved from https://sites.google.com/site/bertrandachou/research
Aldridge, M. D., & Kelley, A.S. (2015). The myth regarding the high cost of end-of-life care. American
Journal of Public Health, 105(12), 2411–2415.
Ameriks, J., Caplin, A., Laufer, S., & Van Nieuwerburgh, S. (2011). The joy of giving or assisted living?
Using strategic surveys to separate public care aversion from bequest motives. Journal of
Finance, 66, 519–561.
Ameriks, J., Briggs, J., Caplin, A., Shapiro, M. D., & Tonetti, C. (2017). Long-term-care utility and late
in-life saving (NBER Working Paper No. 20973). Cambridge, MA: National Bureau of Economic
Research. Retrieved from the National Bureau of Economic Research:
https://www.nber.org/papers/w20973
Ameriks, J., Briggs, J., Caplin, A., Shapiro, M.D., & Tonetti, C. (2018). The long-term-care insurance
puzzle: Modeling and measurement (NBER Working Paper No. 22726). Cambridge, MA: National
Bureau of Economic Research. Retrieved from the National Bureau of Economic Research:
https://www.nber.org/papers/w22726
Baile, W., Lenzi, R., Parker, P., Buckman, R., & Cohen, L. (2002). Oncologists' attitudes toward and
practices in giving bad news: An exploratory study. Journal of Clinical Oncology: Official Journal
of the American Society of Clinical Oncology, 20, 2189-2196.
Baird, M., Hurd, M., & Rohwedder, S. (2014). Medicaid spend-down: The importance of strategic
asset transfers to reach Medicaid eligibility.
Barczyk, D. & Kredler, M. (2018). Long-Term Care across Europe and the U.S.: The Role of Informal
and Formal Care. Available at: www.eco.uc3m.es/mkredler/
Barczyk, D., & Kredler, M. (2017). Evaluating long-term-care policy options, taking the family
seriously. The Review of Economic Studies, 85(2), 766-809.
Bassett, W.F. (2007). Medicaid’s nursing home coverage and asset transfers. Public Finance Review,
35, 414-439.
Bekelman, J.E., Halpern, S.D., Blankart, C.R., Bynum, J.P., Cohen, J., Fowler, R., ... & Oosterveld-Vlug,
M. (2016). Comparison of site of death, health care utilization, and hospital expenditures for
patients dying with cancer in 7 developed countries. The Journal of the American Medical
Association, 315(3), 272-283.
Black, B., Espín-Sánchez, J., French, E., & Litvak, K. (2017). The effect of health insurance on nearelderly health and mortality. American Journal of Health Economics, 3(3), 281-311.
Blau, D.M., & Gilleskie, D. (2008). The role of retiree health insurance in the employment behavior of
older men. International Economic Review, 49(2), 475–514.
Blundell, R., Crawford, R., French, E. & Tetlow, G. (2016). Retirement wealth on both sides of the
pond. Fiscal Studies, 37(1), 105-130.
Bodnar, V. (2016). Insurer in-force long-term care insurance management. In Nordman, E.D. (Ed.),
The State of Long-Term Care Insurance: The Market, Challenges and Future Innovations (pp.

31

71-81). National Association of Insurance Commissioners & the Center for Insurance Policy
Research.
Braun, R.A., Kopecky, K.A. & Koreshkova, T. (2017). Old, sick, alone, and poor: A welfare analysis of
old-age social insurance programmes. Review of Economic Studies, 84(2), 580-612.
Braun, R.A., Kopecky, K.A., & Koreshkova, T. (2018). Old, Frail, and Uninsured: Accounting for
Features of the U.S. Long-term Care Insurance Market (Working Paper 2017-3c). Atlanta, GA:
Federal Reserve Bank of Atlanta. Retrieved from
https://www.frbatlanta.org/research/publications/wp/2017/03c-old-frail-and-uninsuredaccounting-for-puzzles-in-the-us-long-term-care-insurance-2018-09-10.aspx
Brook, R., Ware, J.E., Rogers, W.H., Keeler, E.B., Davies, A.R., Donald, C.A., … & Newhouse, J.P.,
(1983). Does free care improve adults’ health? Results from a randomized trial. New England
Journal of Medicine, 309(23), 1426–1434.
Brown, J.R., & Finkelstein, A. (2008). Medicaid and the long-term care insurance market. American
Economic Review, 98(3), 1083–1102.
Brown, J.R., & Finkelstein, A. (2009). The private market for long‐term care insurance in the United
States: A review of the evidence. Journal of Risk and Insurance, 76(1), 5-29.
Brown, J.R., & Finkelstein, A. (2011). Insuring long-term care in the United States. Journal of
Economic Perspectives, 25(4), 119–142.
Buttorff, C., Ruder, T., & Bauman, M. (2017). Multiple chronic conditions in the United States. Santa
Monica, CA: RAND Corporation, 2017. https://www.rand.org/pubs/tools/TL221.html
Card, D., Dobkin, C., & Maestas, N. (2009). Does Medicare save lives? The Quarterly Journal of
Economics, 124(2), 597–636.
Carr, D., & Luth, E.A. (2017). Advance care planning: Contemporary issues and future directions.
Innovation in Aging, 1(1). Retrieved from https://doi.org/10.1093/geroni/igx012
Centers for Disease Control and Prevention, National Center for Health Statistics. (2018a). Health,
United States 2016 - individual charts and tables: Spreadsheet, PDF, and PowerPoint files.
Retrieved from https://www.cdc.gov/nchs/hus/contents2016.htm
Centers for Disease Control and Prevention. (2018b) National Vital Statistics System [Unpublished
raw data, mortality trend table 290A for 1900-1939 and 1968-78].
Centers for Disease Control and Prevention, National Center for Health Statistics (2018c). Top five
causes of death, 1900, 1950 and 2000. Retrieved from https://data.cdc.gov/NCHS/NCHS-TopFive-Leading-Causes-of-Death-United-State/mc4y-cbbv
Centers for Disease Control and Prevention, National Center for Health Statistics (2018d). Table
47: Current cigarette smoking among adults aged 18 and over, by sex, race, and age: United
States, selected years 1965–2016. Retrieved from
https://www.cdc.gov/nchs/hus/contents2017.htm#047
Centers for Disease Control and Prevention, National Center for Health Statistics. (2018e).
Underlying cause of death 1999-2016. Retrieved from CDC WONDER Online Database:
http://wonder.cdc.gov/ucd-icd10.html
Chernew, M., Cutler, D.M., Ghosh, K., & Landrum, M.B. (2017). Understanding the improvement in
disability-free life expectancy in the US elderly population. In D.A. Wise (Ed.), Insights in the
economics of aging (pp. 161). Chicago, IL: University of Chicago Press.
Cohen, M.A., Kaur, R., & Darnell, B. (2013). Exiting the market: Understanding the factors behind
carriers’ decision to leave the long-term care insurance market. Retrieved from
32

https://aspe.hhs.gov/report/exiting-market-understanding-factors-behind-carriers-decisionleave-long-term-care-insurance-market
Commission on Funding of Care and Support. (2011). Fairer care funding: The report of the
Commission on Funding of Care and Support. Retrieved from:
http://webarchive.nationalarchives.gov.uk/20130221121529/www.wp.dh.gov.uk/carecommissi
on/files/2011/07/Fairer-Care-Funding-Report.pdf
Connor, S., Pyenson, B., Fitch, K., Spence, C., & Iwasaki, K. (2007). Comparing hospice and
nonhospice patient survival among patients who die within a three-year window. Journal of Pain
and Symptom Management, 33, 238-246.
Crimmins, E.M., & Beltrán-Sánchez, H. (2011). Mortality and morbidity trends: Is there compression
of morbidity? The Journals of Gerontology Series B: Psychological Sciences and Social Sciences,
6(1), 75-86.
Cutler, D.M. (2005). Intensive medical technology and the reduction in disability. In D.A. Wise (Ed.),
Analyses in the economics of aging (pp. 161-184). Chicago, IL: University of Chicago Press.
Davidoff, T. (2010). Home equity commitment and long-term care insurance demand. Journal of
Public Economics, 94, 44–49.
Davis M.A., Nallamothu B.K., Banerjee, M., & Bynum, J.P. (2016). Identification of four unique
spending patterns among older adults in the last year of life challenges standard assumptions.
Health Affairs, 35(7), 1316–1323.
De Nardi, M., French, E., & Jones, J.B. (2010). Why do the elderly save? The role of medical expenses.
Journal of Political Economy, 118(1), 37-75.
De Nardi, M., French, E., Gooptu, A., & Jones, J.B. (2012). Medicaid and the elderly. Economic
Perspectives, 36(1), 17-34.
De Nardi, M., French, E., & Jones, J.B. (2016a). Savings after retirement: A survey. Annual Review of
Economics, 8, 177-204.
De Nardi, M., French, E., & Jones, J.B. (2016b). Medicaid insurance in old age. American Economic
Review, 106(11), 3480-3520.
De Nardi, M., French, E., Jones, J.B., McCauley, J. (2016c). Medical spending on the U.S. elderly.
Fiscal Studies, 37(3-4), 327-344.
De Nardi, M., French, E., Jones, J.B., McGee, R. (2018). Couples and singles' savings after retirement.
In progress.
Derby, C.A., Katz, M.J., Lipton, R.B., & Hall, C.B. (2017). Trends in dementia incidence in a birth
cohort analysis of the Einstein Aging Study. JAMA Neurology, 74(11), 1345-1351.
Dobrescu, L. (2015). To love or to pay: Savings and health care in older age. Journal of Human
Resources, 50(1), 254-299.
Doyle, J. (2011). Returns to local-area healthcare spending: Using health shocks to patients far from
home. American Economic Journal: Applied Economics, 3(3), 221–243.
Dynan, K.E., Skinner J., & Zeldes, S.P. (2002). The importance of bequests and life-cycle saving in
capital accumulation: A new answer. American Economic Review, 92, 274–278.
Einav, L., Finkelstein, A., Mullainathan, S., & Obermeyer, Z. (2018). Predictive modelling of U.S.
health care spending later in life. Science, 29, 1462-1465.
Emanuel, E.J., & Emanuel, L.L. (1994). The economics of dying: The illusion of cost savings at the end
of life. New England Journal of Medicine, 330(8), 540-544.
33

Fang, H. (2016). Insurance markets for the elderly. In J. Piggott & A. Woodland (Eds.), Handbook of
the Economics of Population Aging (pp. 237-309). Amsterdam: North Holland Publishing.
Finkelstein, A., & McGarry, K. (2006). Multiple dimensions of private information: Evidence from the
long-term care insurance market. American Economic Review, 96(4), 938-958.
Finkelstein, A., & McKnight, R. (2008). What did Medicare do? The initial impact of Medicare on
mortality and out of pocket medical spending. Journal of Public Economics, 92(7), 1644-1669.
Finkelstein, A., Taubman, S., Wright, B., Bernstein, M., Gruber, J., Newhouse, J.P., … & Oregon Health
Study Group. (2012). The Oregon health insurance experiment: Evidence from the first year.
Quarterly Journal of Economics, 127(3), 1057-1106.
Fisher, E.S., Wennberg, D.E., Stukel, T.A., Gottlieb, D.J., Lucas, F.L., & Pinder, E.L. (2003). The
implications of regional variations in Medicare spending, Part 2: Health outcomes and
satisfaction with care. Annals of Internal Medicine, 138(4), 288-322.
French, E., Baker, O., Doctor, P., DeNardi, M., & Jones, J.B. (2006). Right before the end: New
evidence on asset decumulation at the end of the life cycle. Economic Perspectives, 30(3), 2-13.
French, E., & Kelly, E. (2016). Medical spending around the world: Summary of results. Fiscal Studies,
37(3-4), 717-747.
French, E., Jones, J.B., & McCauley, J. (2017a). The accuracy of economic measurement in the health
and retirement study. Forum for Health Economics and Policy, 20(2).
French, E., McCauley, J., Aragon, M., Bakx, P., Chalkley, M., Chen, S.H., … & Kelly, E.(2017b). Data
from the U.S. and eight other developed countries show that end-of-life medical spending is
lower than previously reported. Health Affairs, 36(7), 1211-1217.
Fries, J.F. (1980). Aging, natural death, and the compression of morbidity. New England Journal of
Medicine, 303, 130-135.
Gawande, Atul. (2014). Being mortal: Medicine and what matters in the end. New York:
Metropolitan Books.
Gonzalo P., Teno J.M. , Mitchell S.L., Skinner J., Bynum J., Tyler D., & Mor, V. (2011). End-of-life
transitions among nursing home residents with cognitive issues. New England Journal of
Medicine, 365(13), 1212-1221.
Gruenberg, E.M. (1977). The failures of success. The Milbank Memorial Fund Quarterly, 55(1), 3-24.
Hackmann, M.B. (2017). Incentivizing better quality of care: The role of Medicaid and competition in
the nursing home industry (NBER Working Paper No. 24133). Cambridge, MA: National Bureau of
Economic Research. Retrieved from the National Bureau of Economic Research:
https://www.nber.org/papers/w24133
Hales, C.M., Fryar, C.D., Carroll, M.D., Freedman, D.S., & Ogden, C.L. (2018). Trends in obesity and
severe obesity prevalence in US youth and adults by sex and age, 2007-2008 to 2015-2016.
Journal of the American Medical Association, 319(16), 1723-1725.
Hendren, N. (2013). Private information and insurance rejections. Econometrica, 81, 1713-1762.
Hogan, C., Lunney, J., Gabel, J., & Lynn, J. (2001). Medicare beneficiaries’ costs of care in the last year
of life. Health Affairs, 20(4), 188-195.
Hoover D.R., Crystal S., Kumar R., Sambamoorthi U., & Cantor J.C. (2002). Medical expenditures
during the last year of life: Findings from the 1992-1996 Medicare current beneficiary survey.
Health Services Research, 37(6), 1625-1642.

34

Howdon, D. & Rice, N. (2018). Health care expenditures, age, proximity to death and morbidity:
implications for an ageing population. Journal of Health Economics, 57, 60-74.
Kelley, A.S., & Bollens-Lund, E. (2018). Identifying the population with serious illness: The
“denominator” challenge. Journal of Palliative Medicine, 21(S2), S-7.
Kelley A.S., McGarry K., Gorges, R., & Skinner J.S. (2015). The burden of health care costs for patients
with dementia in the last 5 years of life. Annals of Internal Medicine, 163(10), 729-736.
Khwaja, Ahmed. (2010). Estimating willingness to pay for Medicare using a dynamic life-cycle model
of demand for health insurance. Journal of Econometrics, 156(1), 130-147.
Ko, Ami, (2018). An equilibrium analysis of the long-term care insurance market. Retrieved from
http://www.ko-ami.com/
Konetzka, R.T., He, D., Guo, J., & Nyman, J.A. (2014). Moral hazard and long-term care insurance.
Retrieved from https://business.illinois.edu/nmiller/mhec/Konetzka.pdf
Kopecky, K., & Koreshkova, T., (2014). The impact of medical and nursing home expenses and social
insurance policies on savings and inequality. American Economic Journal: Macroeconomics, 6,
29-72.
Krakauer, R., Spettell, C.M., Reisman, L. & Wade, M.J., (2009). Opportunities to improve the quality
of care for advanced illness. Health Affairs, 28(5), 1357-1359.
Kvaerner, J. (2018). How Strong Are Bequest Motives? Retrieved from SSRN:
http://dx.doi.org/10.2139/ssrn.2985465
Lockwood, L.M. (2018). Incidental bequests and the choice to self-insure late-life risks. American
Economic Review, 108(9), 2513-2550.
Lubitz, J.D., & Riley, G.F. (1993). Trends in Medicare payments in the last year of life. New England
Journal of Medicine, 328, 1092-1096
Marshall, S., McGarry, K., & Skinner, J. S. (2011). The risk of out-of-pocket health care expenditures
at the end-of-life. In D.A. Wise (Ed.), Explorations in the economics of aging, Chicago, IL:
University of Chicago Press.
Martin, R.S., Hayes, B., Gregorevic, K., & Lim, W.K. (2016). The effects of advance care planning
interventions on nursing home residents: A systematic review. Journal of the American Medical
Directors Association, 17, 4284-4293.
de Meijer, C., Koopmanschap, M., d’Uva, T.B., & van Doorslaer, E. (2011). Determinants of long-term
care spending: age, time to death or disability? Journal of Health Economics, 30(2), 425-438.
Meckel, R.A. (1990). Save the babies: American public health reform and the prevention of infant
mortality, 1850-1929. Baltimore, Maryland: The Johns Hopkins University Press.
Mitchell, S.L., Mor V., Gozalo, P.L., Servadio, J.L., Teno J.M. (2016). Tube feeding in US nursing home
residents with advanced dementia, 2000–2014. Journal of the American Medical Association,
316(7), 769-770.
Mitchell, S.L., Teno J.M., Roy J., Kabumoto, G., Mor, V. (2003). Clinical and organizational factors
associated with feeding tube use among nursing home residents with advanced cognitive
impairment. Journal of the American Medical Association, 290(1), 73-80.
Mommaerts, C. (2016). Long-term care insurance and the family. Retrieved from
https://www.aeaweb.org/conference/2017/preliminary/paper/Ak22d8D9

35

National Hospice and Palliative Care Organization. (2014). NHPCO facts and figures: Hospice care in
America (2014 Edition). Retrieved from
https://www.nhpco.org/sites/default/files/public/2014_Facts_Figures.pdf
National Hospice and Palliative Care Organization. (2018). NHPCO facts and figures: Hospice care in
America (2017 Edition). Retrieved from
https://www.nhpco.org/sites/default/files/public/Statistics_Research/2017_Facts_Figures.pdf
Norton, E.C. (1995). Elderly assets, Medicaid policy, and spend‐down in nursing homes. Review of
Income and Wealth, 41(3), 309-329.
Nakajima, M., & Telyukova, I.A. (2016). Housing and saving in retirement across countries. In J.
Stiglitz & M. Guzman (Eds.), Contemporary issues in microeconomics (pp. 88-126). London:
Palgrave Macmillan.
Nakajima, M., & Telyukova, I.A. (2017). Reverse mortgage loans: A quantitative analysis. Journal of
Finance, 72(2), 911-950.
Nakajima, M., & Telyukova, I.A. (2018). Medical expenses and saving in retirement: The case of U.S.
and Sweden (Institute Working Paper 8.) Minneapolis, MN: Federal Reserve Bank of
Minneapolis. Retrieved from https://www.minneapolisfed.org/institute/institute-workingpapers/medical-expenses-and-saving-in-retirement-the-case-of-us-and-sweden
National Cancer Institute. (2017). Annual report to the nation 2017: Special section: Survival.
Retrieved from https://seer.cancer.gov/report_to_nation/survival.html
The Organisation for Economic Co-operation and Development. (2018). Health expenditure and
financing. Retrieved from https://stats.oecd.org/Index.aspx?DataSetCode=SHA
Orlovic, M., Marti, J., & Mossialos, E. (2017). Analysis of end-of-life care, out-of-pocket spending, and
place of death in 16 European countries and Israel. Health Affairs, 36(7), 1201-1210.
Oster, E., Shoulson, I., Quaid, K., & Dorsey, E.R. (2010). Genetic adverse selection: Evidence from
long-term care insurance and Huntington disease. Journal of Public Economics, 94(11-12),
1041-1050.
Pauly, M.V. (1990). The rational nonpurchase of long-term-care insurance. Journal of Political
Economy 98(1), 153-168.
Payne, G., Laporte, A., Deber, R., & Coyte, P.C. (2007). Counting backward to health care's future:
using time-to-death modeling to identify changes in end-of-life morbidity and the impact of
aging on health care expenditures. The Milbank Memorial Fund Quarterly, 85(2), 213-257.
Poterba, J., Venti, S., & Wise, D. (2011). The composition and drawdown of wealth in retirement.
Journal of Economic Perspectives, 25(4). 95-118.
Public Health England. (2017a). Changes in children’s body mass index between 2006/07 and
2015/16: National Child Measurement Programme. Retrieved from
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data
/file/646271/national_child_measurement_programme_changes_in_childrens_BMI_detailed_re
port.pdf
Public Health England. (2017b). Life expectancy and healthy life expectancy. In Health profile for
England. Retrieved from https://www.gov.uk/government/publications/health-profile-forengland
Quaresma, Manuela, Coleman, M.P., & Rachet, B. (2015). 40-year trends in an index of survival for
all cancers combined and survival adjusted for age and sex for each cancer in England and
Wales, 1971–2011: A population-based study. The Lancet, 385, 1206-1218.
36

Reid, T.R. (2017). How we spend $3,400,000,000,000. The Atlantic. Retrieved from
https://www.theatlantic.com/health/archive/2017/06/how-we-spend-3400000000000/530355/
Riley, G.F., & Lubitz, J.D. (2010). Long-term trends in Medicare payments in the last year of life.
Health Services Research, 45(2), 565-576.
Rodrigues, R., Huber, M., & Lamura, G. (Eds.). (2012). Facts and figures on healthy ageing and
long-term care. Europe and North America, Occasional Reports Series 8. Vienna: European
Centre.
Scitovsky, A.A. (2005). The high cost of dying: What do the data show? The Milbank Memorial Fund
Quarterly, 83(4), 825-841.
Seshamani, M., & Gray, A.M. (2004). A longitudinal study of the effects of age and time to death on
hospital costs. Journal of Health Economics 23(2), 217-235.
Silveira, M.J., Wiitala, W., & Piette, J. (2014). Advance directive completion by elderly Americans: A
decade of change. Journal of the American Geriatrics Society, 62(4), 706-710.
Singer, P.A., Martin, D.K. & Kelner, M. (1999). Quality end-of-life care: Patients' perspectives. Journal
of the American Medical Association, 281(2), 163-168.
Spettell, C.M., Rawlins, W.S., Krakauer, R., Fernandes, J., Breton, M.E.S., Gowdy, W., … & Brennan,
T.A. (2009). A comprehensive case management program to improve palliative care. Journal of
Palliative Medicine, 12(9), 827-832.
Stadtmauer, E.A., O'Neill, A., Goldstein, L.J., Crilley, P.A., Mangan, K.F., Ingle, J.N., … & Sickles, C.,
(2000). Conventional-dose chemotherapy compared with high-dose chemotherapy plus
autologous hematopoietic stem-cell transplantation for metastatic breast cancer. New England
Journal of Medicine, 342(15), 1069-1076.
Stafford, M., Steventon, A., Thorlby, R., Fisher, R., Turton, C., & Deeny, S. (2018). Understanding the
health care needs of people with multiple health conditions. Retrieved from
https://www.health.org.uk/publications/understanding-the-health-care-needs-of-people-withmultiple-health-conditions
Steinhauser, K.E., Christakis N.A., Clipp E.C., McNeilly M., McIntyre L., & Tulsky, J.A. (2000). Factors
considered important at the end of life by patients, family, physicians, and other care
providers. Journal of the American Medical Association, 284(19), 2476–2482.
doi: 10.1001/jama.284.19.2476
Taylor, D.H. Jr., Sloan, F.A., & Norton, E. C. (1999). Formation of trusts and spend down to Medicaid.
The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 54, 194-201.
Temel, J.S., Greer, J.A., Muzikansky, A., Gallagher, E.R., Admane, S., Jackson, V.A., … & Lynch, T.J.
(2010). Early palliative care for patients with metastatic non-small-cell lung cancer. The New
England Journal of Medicine, 363, 733-742.
Teno, J.M., Gozalo, P.L., Bynum, J.P., Leland, N.E., Miller, S.C., Morden, N.E., … & Mor, V. (2013).
Change in end-of-life care for Medicare beneficiaries: Site of death, place of care, and health
care transitions in 2000, 2005, and 2009. Journal of the American Medical Association, 309(5),
470-477.
Teno, J.M., Gruneir, A., Schwartz, Z., Nanda, A., & Wetle, T. (2007). Association between advance
directives and quality of end-of-life care: A national study. Journal of the American Geriatrics
Society, 55(2), 189-194.
Teno, J.M., Lynn, J., Wenger, N., Phillips, R.S., Murphy D.P., & Connors, A.F. (1997). Advance
directives for seriously ill hospitalized patients: Effectiveness with the Patient Self-Determination
Act and the SUPPORT intervention: SUPPORT investigators: Study to understand prognoses and
37

preferences for outcomes and risks of treatment. Journal of the American Geriatrics Society,
45(4), 500-507.
U.S. Centers for Medicare & Medicaid Services. (2018). Your Medicare coverage: Skilled nursing
facility (SNF) care. Retrieved from https://www.medicare.gov/coverage/skilled-nursing-facilitysnf-care
Waidmann, T., & Liu, K. (2006). Asset transfer and nursing home use: Empirical evidence and policy
significance (Issue Paper 7847). Retrieved from the Kaiser Commission on Medicaid and the
Uninsured, Kaiser Family Foundation: https://www.kff.org/medicaid/issue-brief/asset-transferand-nursing-home-use-empirical
Wiener, J.M., Anderson, W.L., Khatutsky, G., Kaganova, Y., & O’Keeffe, J. (2013). Medicaid spend
down: New estimates and implications for long-term services and supports financing reform.
Retrieved from https://www.rti.org/sites/default/files/resources/tsf_ltc-financing_medicaidspend-down-implications_wiener-tumlinson_3-20-13_0.pdf
Will, G.F., (1990). John Silber and the limitations on resources. Baltimore Sun. Retrieved from
http://articles.baltimoresun.com/1990-09-23/topic/0406090447_1_john-silber-terminally-illresources
Wright, A.A., Zhang, B., Ray, A., Mack, J.W., Trice, E.D., Balboni, T.A., … & Prigerson, H.G. (2008).
Associations between end-of-life discussions, patient mental health, medical care near death,
and caregiver bereavement adjustment. Journal of the American Medical Association, 300(14),
1665-1673.
Xu, J.Q., Murphy, S.L., Kochanek, K.D., Bastian, B. & Arias, E. (2018). Deaths: Final data for 2016.
National Vital Statistics Reports, 67(5). Hyattsville, MD: National Center for Health Statistics.
2018.
Yang, Z., Norton, E.C., & Stearns, S.C. (2003). Longevity and health care expenditures the real
reasons older people spend more. The Journals of Gerontology Series B: Psychological Sciences
and Social Sciences, 58(1), S2-S10.
Zweifel, P., Felder, S., & Meiers, M. (1999). Ageing of population and health care expenditure: A red
herring? Health Economics, 8(6), 485-496.

38