ArticlePDF AvailableLiterature Review

Abstract

Hardly a week goes by without a front-page newspaper article on rising health care costs and the uninsured. In this article, I focus mainly on costs, arguing that the issue has been somewhat misconceived: while the level of medical care spending in the U.S. is a cause for concern, the welfare losses associated with rises in that level of spending may not be as large as the public rhetoric can make them seem. In fact, cost containment may not be as urgent as is widely supposed, and some proposed "cost containment" policies may result in welfare losses for the insured and even increase the number of uninsured
Journal of
Economic Perspectives
Volume
6, Number 3
Summer
1992
Pages
3–21
Medical Care Costs: How Much
Welfare Loss?
Joseph P. Newhouse
H
ardly a week goes by without a front-page newspaper article on rising
health care costs and the uninsured. In this article, I focus mainly on
costs,
arguing that the issue has been somewhat misconceived: while
the level of medical care spending in the U.S. is a cause for concern, the welfare
losses associated with rises in that level of spending may not be as large as the
public rhetoric can make them seem. In fact, cost containment may not be as
urgent as is widely supposed, and some proposed "cost containment" policies
may result in welfare losses for the insured, and even increase the number of
uninsured.
The Magnitude of the Medical Expenditure Increase
As both newspapers and pay stubs remind us, the health care sector is a
large part of the economy; spending on health care, $666 billion in 1990,
exceeded 12 percent of GNP. At the federal level, Medicare (for the elderly)
and Medicaid (for the poor) represent 15 percent of outlays. Medicare is the
second largest domestic program of the federal government. At the state level,
Medicaid is 11 percent of total expenditures (1991 Economic Report of the
President, Tables B-76, B-82).
Despite these magnitudes, I think the crux of public concern over health
care costs comes from the data in Table 1; real medical care expenditures per
capita have been growing at around 4 percent per year for five
decades,
except
Joseph P. Newhouse is John D. MacArthur Professor of Health Policy and Manage-
ment, Harvard University, Cambridge, Massachusetts.
4 Journal of Economic Perspectives
Table 1
Growth in Real Health Care Expenditure and GNP, by Decade
(% per year)
in the 1960s when growth was even more rapid. At these rates of growth, the
share of GNP devoted to medical care has also been steadily growing.
Neither citizens nor economists, of course, are especially concerned about
rapid growth in most sectors of the economy, like the personal computer
industry, the fax machine industry, or the cellular phone industry. The conven-
tional explanation of why growth in medical costs is different emphasizes the
moral hazard from health insurance and particularly the tax treatment of
health insurance.
Traditional health insurance reimburses as a function of expenditure or
use.
Because insurance drives the marginal price of medical care at the point of
use to near zero, so the usual view goes, consumers—or physicians acting as
their agents—demand care until the marginal product of additional care is
nearly zero. To use Alain Enthoven's (1980) phrase, we engage in "flat-of-the-
curve" medicine, where spending on medical care increases even though the
additional gains from such spending are very low or nonexistent.
1
Moreover,
the argument continues, the exclusion of employer-paid premiums from tax-
able income exacerbates this situation by leading to excessive health insurance
(Feldstein and Allison, 1974). Pauly (1986) offers a review of this literature; a
1
This view is buttressed by studies showing a reasonably high percentage of procedures—a sixth to
a third, among three procedures studied—that upon clinical review were deemed to be "medically
inappropriate" (Chassin et al., 1987).
Joseph P. Newhouse 5
more recent example of this line of argument is the 1991
Economic
Report of
the
President (p. 142).
However, I will contend that economists have been too preoccupied with a
one-period model of health care services that takes technology as given, and
that we need to pay more attention to technological change.
2
In particular, the
literature has focused on the tradeoff between moral hazard and risk sharing in
the context of the one-period model. The cocktail party story of excessive
medical spending built around that model has some validity, of course. The
price elasticity of demand for medical care services is not zero. Far from it; a
fully insured population spends about 40 to 50 percent more than a population
with a large deductible, and their health status is not measurably improved by
the additional services (Manning et al., 1987).
In this essay I wish to distinguish the welfare losses at a point in time from
those that may occur because of the increases in expenditure over time. My
initial focus, then, is why medical expenditure is increasing, as opposed to
merely "high." To explain increasing expenditure, one needs to point to
something that is changing, indeed to factors that have been changing for 50
years.
Accounting for the Medical Expenditure Increase
In accounting for the expenditure increase, my methods will be similar to
those in the literature accounting for economic growth; identify a series of
factors, determine how much of the change they might account for, and
attribute the residual to technological change. In particular, I consider three
factors known to affect demand for medical services—aging, the spread of
insurance, and the growth of income—as well as two factors that are said to
affect the supply of services—supplier-induced demand and differential pro-
ductivity growth. By deferring consideration of technological change, I want to
establish that the one-period model, with constant technology, cannot give us
much help in explaining the expenditure increase.
Aging
The proportion of elderly has been growing steadily for 50 years, which
might be expected to raise expenditures on medical care. But by how much?
One way to tackle this question would be to calculate by how much spending
would increase solely as a result of the changing proportion by age group, if
spending per person in each age group were fixed over time. For example, the
proportion of the population between the ages of 19 and 64 stayed nearly
constant between 1950 and 1987, rising from 59 to 60 percent. However, the
2
Goddeeris (1984a, 1984b) is a notable exception, as is Weisbrod's (1991) survey article of a year
ago.
6 Journal of Economic Perspectives
population of those over 65 increased from 8 percent of the population in 1950
to 12 percent in 1987, while the proportion of those below age 19 fell from 33
percent to 28 percent. Using the weights given by the average expenditures per
person in each age group in 1987, this swing can account for an increase of 15
percent in total spending.
3
However, the data in Table 1 imply that real
expenditure per capita grew over this period by more than a factor of 5. Thus,
aging can only account directly for a tiny fraction of the increase in expendi-
ture.
Victor Fuchs (1990) has questioned this type of calculation, pointing out
that other effects may be associated with aging. He notes that spending rises
rapidly in the period shortly before death and that as age-specific mortality
rates fall, fewer people at each age are close to death. This would cause the
method just described to understate the effect of aging on costs, essentially
because deaths are concentrated among the elderly, so spending on the elderly
relative to the non-elderly is temporarily depressed while mortality rates are
falling. To get some sense of the quantitative importance of this argument, I
calculated the effect of aging using elderly to non-elderly spending rates from
1977 rather than 1987.
4
The 1977 values give virtually the same result as those
from
1987.
Thus, this criticism does not seem quantitatively important.
Fuchs also speculates that falling age-specific mortality rates may increase
the propensity to intervene medically at any age, which provides an alternate
reason why an aging population may consume ever more medical care. It is
more difficult to assess this argument quantitatively, in part because causation
could run the other way; an increased propensity to intervene could lower the
age-specific mortality rate. However, whichever way the causation runs, the
magnitude of the negative correlation between mortality and spending may not
be large. In the four previous decades, age-adjusted death rates fell least in the
1960s and most in the 1970s, but the rate of expenditure increase was greatest
in the 1960s and next to least in the 1970s. Unfortunately, the 1960s also saw
the introduction of Medicare and Medicaid, which makes any inference from
these data about Fuchs' argument problematic. On balance, however, neither
the direct nor indirect effects of aging on expenditure appear to account for
much of the sustained rise in medical expenditure.
Increased Insurance
This explanation is probably closest to the conventional wisdom among
economists; the spread of insurance has steadily reduced price to the consumer
3
I used the 1987 data for this calculation because I have data on medical spending by five-year age
group for the over-65 for that year and thus can better adjust for aging within the over 65 group.
The data are from Waldo et al. (1989) with corrections supplied by Daniel Waldo. Also, I use the
period beginning in 1950 because I could not readily locate 1940 data on age, but there can be
little doubt that the conclusion would be the same.
4
The calculation for 1977 is somewhat cruder. I only have data on medical spending per person for
the under 19, 19–64, and over 65 age groups for 1977, so I could not use the five-year age cohorts
as described in the previous note. In comparing 1977 and 1987 results, I have used the spending
comparisons for these three age groups in both cases.
Medical Care Costs: How Much Welfare Loss? 7
and driven up demand for medical services, thereby resulting in a steady
expenditure increase. Like the aging argument, this argument has an element
of truth, but it also leaves the great bulk of the increase unaccounted for.
The data from Manning et al. (1987) already referred to show that the
effect of moving from an average coinsurance rate of 33 percent to a coinsur-
ance rate of zero at a point in time is roughly a 40 to 50 percent increase in
demand.
5
The change in the average coinsurance rate from 1950 to 1980 was
from 67 percent to 27 percent. Thus, if a function relating demand to the
average coinsurance rate were linear and if there were no technological change,
the 40 percentage point drop in coinsurance should have caused about a 50
percent increase in demand from 1950 to 1980.
6
Thus, the factor-of-five
increase in real expenditure per person over this period is perhaps eight times
as large as one could account for solely from the effect of increased insurance
on demand in the context of the one-period model.
Another fact that is not consistent with the importance of increased de-
mand from the spread of insurance is that the largest single component of the
increase in real medical expenditure per person has been in the hospital sector.
Yet the average coinsurance rate for hospital services was essentially constant at
about 5 percent in the 1980s (Levit et al., 1991) while real hospital expenditure
rose over 50 percent during the decade.
Increased Income
Medical care is a normal good, so growth in income can also account for an
increase in expenditure. Estimates of the income elasticity of demand for
medical care within the United States (done with cross-section observations
across households, holding insurance constant) are around 0.2–0.4. From 1940
to 1990, real GNP per capita increased by 180 percent. Using an estimate of
income elasticity of 0.2 to 0.4, income growth (given technology) could account
for around a 35 to 70 percent increase in expenditure, obviously a small
portion of the 780 percent actual increase.
In international cross sections of developed countries, with the country as
the unit of observation, however, income elasticities of demand for medical care
are around 1.0 or even more (Newhouse, 1977; Parkin, McGuire, and Yule,
1987;
Gerdtham et al., 1992). Using aggregate relationships can lead to well-
known problems (Parkin, McGuire, and Yule, 1987). Nonetheless, one might
5
Estimated price elasticities are in the range of 0.1 to 0.2. See Manning et al. (1987) for three
different methods of calculating price elasticities.
6
There are serious conceptual problems in relating demand to an average coinsurance rate; for
example, an average coinsurance rate that comes from a deductible followed by full coverage is
likely to have different effects than from a constant coinsurance rate. Thus, one should not expect
the function relating an average coinsurance rate to medical expenditure to be unique. Nonethe-
less,
I am only trying to get a rough feel for the importance of various factors, and for that purpose
ignoring the conceptual problems associated with this function seems permissible. Moreover, it is
worth noting that if the relevant function was constant elasticity rather than linear, there would be
a smaller increase in demand than that estimated in the text. For a discussion of these problems
and their implications for estimation of demand, see Newhouse, Phelps, and Marquis (1980).
8 Journal of Economic Perspectives
prefer this estimate of the income elasticity because the within-country income
elasticities may be distorted by the endogeneity of income at the household
level; sickness may simultaneously depress income and raise medical spending.
Using 1.0 as an income elasticity rather than 0.2 to 0.4, one could account for a
little under a quarter of the overall increase.
7
Supplier-Induced Demand
A substantial literature in health economics has been concerned with
supplier- or physician-induced demand (Cromwell and Mitchell, 1986; Rice,
1983).
This literature has not reached consensus on the magnitude of supplier-
induced demand, but even granting its importance at a point in time, one can
ask about its importance over time. That is, if physicians have considerable
discretion in treating ignorant consumers, to what degree might they have
induced more and more demand?
Those who emphasize supplier-induced demand as a factor in the expendi-
ture increase argue that as physician supply has grown, physicians have in-
creased demand to protect their incomes (for the theory underlying this view
see McGuire and Pauly, 1991). The evidence, however, does not offer much
support to the view that supplier-induced demand is important in the rate of
change. As a crude measure, I list the decade-by-decade changes in physician
supply per person in Table 2. There is no correlation with the similar data on
expenditure given earlier in Table 1.
I do not want to make too much of the lack of a simple correlation among
six observations. One could, for example, argue that economy-wide growth has
slowed since 1970, decreasing demand (ceteris paribus) and offsetting induce-
ment effects from the increase in physician supply. Nonetheless, the lack of any
obvious change in the rate of expenditure growth after 1970, when physician
supply clearly increased, is striking.
A variant of the supplier-induced demand argument is "defensive
medicine," or the use of tests and procedures with little or no value to patients
to minimize the chance of a successful malpractice suit. Although the amount of
such medicine is uncertain, the most widely cited estimate pegged it at around
1 percent of all medical expenditure in 1984 (Reynolds, Rizzo, and Gonzalez,
7
Income elasticities from time series data within countries are typically around 1. For example,
Schieber and Poullier (1989) present data for 20 OECD countries from 1975 to 1987. The countries
have income elasticities that vary from 0.9 to 1.3, with 10 having a value of 1.1. (I use the
"nominal" income elasticities shown in their Table 2 because I believe there are conceptual
problems with the health price deflator they use to estimate "real" elasticities. See the text below.)
In general one might expect that time series income elasticities would exceed the cross section
income elasticities; technology is not held constant in the time series, and the income elasticity for
new technology should be positive, thus adding to the income elasticity for additional care given
technology observed at a point in time. Because of the magnitude of the time-series income
elasticity, I am inclined to believe that the income elasticity relevant to the calculation in this
section, which is the income elasticity for additional care given technology, would be well under 1.
Joseph P. Newhouse 9
Table 2
Decade-by-Decade Growth in Numbers
of Physicians per Person
(annual rate of increase)
1987).
Thus, even if defensive medicine were zero in 1940, its growth can only
account for a trivial fraction of the expenditure increase.
Factor Productivity in a Service Industry
Yet another argument that has been made to account for increases in
medical spending is that medical care is a service. If productivity gains are
lower for services like medical care than in the rest of the economy, then
relative medical prices would rise over time; because demand is inelastic
expenditures would also rise (Baumol, 1988). The magnitude of productivity
gains in medical care is an exceedingly difficult question because of the
difficulty in measuring the product. Perhaps zero productivity growth is appro-
priate for long-term care and home health care. These sectors, however,
represent only about 10 percent of the entire medical care sector. In the light
of the technological change discussed in the next section, I think the assump-
tion of no productivity gain for acute medical care services is implausible;
indeed, a true productivity measure might even go up at or in excess of
economy-wide rates. For example, the treatment of heart attacks has certainly
changed more than haircuts or the performance of Mozart string quartets,
standard examples of services whose productivity has scarcely changed. Thus, it
is not clear that much of the expenditure increase should be assigned to this
factor.
In principle, a price index for medical care offers the potential to shed
light on this point. If medical care inflation were similar to general price
inflation, one could reject the concept that differential productivity growth is
driving the price levels apart. In the 1980s, medical price inflation exceeded
10 Journal of Economic Perspectives
the general consumer price index by 3.4 percent per year, which would seem
to offer some support for the idea that lagging productivity growth is the
reason more is spent on medical care. However, the gap was only .4 percent
per year in the 1970s, 2.0 percent per year in the 1960s, and 1.9 percent per
year in the 1950s.
Moreover, the consumer price index for medical care has severe measure-
ment problems, so severe so that I would not draw any inferences on productiv-
ity from it. Four main problems include:
First, the consumer price index for medical services focuses on the costs of
physician visits and days in the hospital, not the costs of treating ailments. As
Scitovsky (1967) emphasized a quarter century ago, the product the consumer
is really purchasing is the treatment of a medical problem. The relevant cost,
therefore, is the cost of treating, say, a heart attack or appendicitis, not the cost
of a day in the hospital. Especially in the 1980s, length of stay for various
diseases has been falling sharply. Setting aside the issue of whether this reflects
a change in quality, a true price index would need to reflect the total cost of
treatment, but the existing index does not. Moreover, the last day(s) of a stay
are less resource-intensive than the first days. Thus, a fall in the average length
of stay will increase the resource intensity per day, which conceivably could
result in an increase in the price per day; that is, the consumer price index
could be signalling an increased price (per day) when there was actually a
decreased price (per treatment). By focusing on the cost of a day in the
hospital, the consumer price index also fails to register any savings when the
treatment of some diseases, like cataracts, is shifted out of the hospital to
out-patient surgery or when a new drug changes treatment so that surgery is
not required as often, as has happened with ulcers.
Second, the consumer price index has historically used the list price for
medical care, not the actual transaction price. In fact, there are hardly any
transactions at the list price that the CPI uses for hospitals. The Bureau of
Labor Statistics is attempting to correct this problem. But Dranove et al. (1991)
estimate that this factor alone caused the CPI to overstate hospital price
increases in California by 40 percent in the 1980s.
Third, the consumer price index makes no adjustment for quality change.
With the rate of introduction of new products and procedures into medical
care,
surely acting as if quality is unchanged will lead to large overstatement in
the medical portion of the consumer price index.
Fourth, the various components of the medical care portion of the con-
sumer price index—hospital, physician, dental, drug—are combined into an
overall medical care price index using weights proportional to out-of-pocket
expenditure. Because insurance coverage differs between these components,
hospital spending, which accounts for about 40 percent of personal medical
care spending, has a weight in the medical care CPI that is half as large.
Indeed, the weight on hospital spending is only slightly larger than the weight
on dental spending, although dental spending accounts for less than 10 percent
Medical Care Costs: How Much Welfare Loss? 11
of total health care spending. This criticism affects the component price indices
much less, but it certainly affects one's ability to use the overall medical care
CPI to decompose the increase in health care expenditure into increases in
price and increases in quantity, as is needed to test the productivity-based
explanation for growth in overall medical expenditures.
8
For all these reasons I do not believe we have any good empirical basis for
decomposing the medical care expenditure increase into increases in price and
increases in quantity.
Technological Change in Medical Care
Because of the problem in measuring productivity, it is hard to know how
much of the increase all the above factors can account for (assuming no
technological change). My own view is that they account for well under
half—perhaps under a quarter—of the 50-year increase in medical care expen-
diture. Thus, we are left with trying to explain a large residual.
I believe the bulk of the residual increase is attributable to technological
change, or what might loosely be called the march of science and the increased
capabilities of medicine. By technological change I mean not only new types of
physical capital, such as magnetic resonance imaging, but also new procedures,
such as coronary artery bypass grafting. Other examples abound: renal dialysis;
transplantation; artificial joints; endoscopy; monoclonal antibodies; drugs for
mental illness. Both the last example and the polio vaccine example emphasize
that technological change is not necessarily expenditure-increasing, but if it is
on balance expenditure-lowering, as is sometimes argued (Thomas, 1975),
accounting for the rise in expenditure is more difficult than ever.
Trying to attribute a residual to a specific factor is an inherently frustrating
exercise, and the best I can do to support my argument that much of the
residual is attributable to the new capabilities of medicine—new product lines,
if one prefers—is to buttress it with data that I believe are consistent with it.
There are three such sorts of data.
The first are shown in Table 3. If medical technology were constant, then
increased demand from more elderly, more insurance, and more income
should result in more hospital patient days. However, patients are not going to
the hospital more frequently—indeed, admission rates are now barely above
1960 levels and age adjustment would wipe out the difference—nor are
patients staying longer; length of stay has fallen. But the real cost of a day (or a
stay) in the hospital rose by nearly a factor of 4 from 1965 to 1986. (Data are
8
Even the component indices are affected to the degree that the actual services priced have
differential insurance coverage. For example, prior to the late 1970s, insurance commonly did not
cover maternity services, or covered such services with a lump-sum allowance. This caused the
hospital component of the CPI to overweight maternity services if the aim was to decompose
hospital expenditure increases into price and quantity increases.
12 Journal of Economic Perspectives
Table 3
Utilization of Short Stay General Hospitals
not readily available from before 1965.) Thus, what is being done to and for
people who are in the hospital is affecting hospital costs, not an increased
number of people at the hospital.
Remember that the increase in hospital expenditure is the single largest
component of the overall expenditure increase. The evidence that this increase
has occurred in the cost of a patient day, not in the rate of patient days per
person, is certainly consistent with a story that technological change accounts
for the bulk of the increase in costs.
9
The second set of data are data on costs of enrollees in Health Maintenance
Organizations (HMOs). HMOs combine the insurance and delivery functions
and are generally paid a fixed dollar amount per month per enrollee. The costs
borne by enrollees in HMOs have risen at roughly the same rate as personal
health care expenditures, albeit at each point in time they are at a lower level
(Newhouse et al., 1985). Although HMOs appear to eliminate some of the low
benefit (relative to cost) hospitalization at each point in time (Luft, 1981; Sloss
et al., 1987), whatever drives up costs in the predominantly fee-for-service
medical care system also drives up costs in HMOs. The factor or factors
responsible for increasing costs could be common rises in factor prices in both
systems, but common changes in the technology available is also consistent with
the data.
Third, international comparisons also underline the conclusion that dif-
ferent types of medical systems have much the same increases in spending. The
9
Nor have visit rates to physicians much changed. In 1958–59 they were 4.7 per person; in 1970s,
4.6;
in 1980, 4.8; and in 1989, 5.4. Data are from the National Health Interview Survey for the
respective year.
Joseph P. Newhouse 13
annual percent increase in real per capita health spending in the U.S. from
1960 to 1987 was 5.0 percent. The rate of increase was lower in the United
Kingdom, Canada, and Germany, at 3.7, 4.7, and 4.8 percent, respectively. But
the rate of increase was higher in France, Italy, and Japan, at 5.8, 6.0, and 8.9
percent, respectively.
10
In the light of the pronounced institutional differences
among these countries in medical care financing arrangements, the similarity in
the real rate of growth is striking. Thus, whatever is behind the medical care
expenditure rise appears to be common across these countries. The advance in
medical technology is, of course, one such common factor.
Was Technological Change Induced by Insurance?
I come below to some normative comments about the increase in costs.
One's views about the cost increases, however, might differ if one thought the
change was induced by high levels of insurance, perhaps in turn induced by the
tax subsidy, such that there was a welfare loss from too much change. At this
point I only want to point out the fallacy in an argument sometimes heard in
the public debate—that without insurance people could not afford to pay for
much of the technology; hence, the change was induced by insurance, and ipso
facto the technology represents a welfare loss. The essence of the fallacy is that
the appropriate first-best condition is not what would be observed in an
uninsured market, but what consumers are willing to pay for an insurance
policy that would cover the technology in relevant states of the world (Marshall,
1976;
De Meza, 1983; Goddeeris, 1984a, 1984b). In other words, a principal
function of insurance is to transfer income across states of the world; one may
well want more income, for example, in a state of the world where one's
kidneys have failed. The inference that kidney dialysis or kidney transplant
technology necessarily represents a welfare loss because most uninsured con-
sumers could not afford it is thus incorrect.
What Has Been Done About Medical Costs?
Over the years a variety of "cost containment" techniques have been tried.
On balance, these techniques appear to have been beneficial, but they have had
primarily a once-and-for-all effect on the level of medical expenditure, leaving
the steady-state rate of change little affected (Schwartz, 1987). I will discuss
three of the recent efforts to control costs by both the private and the public
sector.
10
The
bulge in Japan is entirely attributable to the decade of the 1960s, when real per capita GDP
in Japan grew by 11 percent per year and real health care expenditure grew by 14 percent per
year, thus helping drive up expenditures on health care along with everything else.
14 Journal of Economic Perspectives
First, initial cost sharing by patients has increased. Private insurers in-
creased deductibles substantially in the early 1980s. For example, the propor-
tion of firms with a deductible of $200 or more in a sample of medium
and large firms rose from 4 percent to 21 percent between 1982 and 1984
(Goldsmith, 1984). Even more strikingly, the number of such firms that had no
initial cost sharing for hospital services fell from 70 percent to 37 percent in
that two-year period. This increased amount of cost sharing almost certainly
was responsible for part of the decline in admissions shown earlier.
11
Although this increase in initial cost sharing appears effective and efficient,
there is little reason to think that it will change the long-term rate of cost
increases. Indeed, another trend in the 1980s, toward caps on out-of-pocket
spending, will increase cost, since it creates a marginal price of zero for most
hospitalized patients.
12
The spread of such caps provides better protection
against risk, but has meant an even greater proportion of hospital patients face
no cost for the marginal service or procedure, consistent with the increase in
services delivered per stay.
Increased enrollment in health maintenance organizations has been a
second approach to cost containment. The federal government has encouraged
enrollment in health maintenance organizations since the mid-1970s as a
cost-containment technique. We have already seen, however, that the effects on
costs were of the once-and-for-all variety; the rate of growth in HMO spending
appears similar to the entire sector.
A third method of addressing costs has been the adoption of prospective
payment systems. In particular, in 1983 Medicare began to reimburse hospitals
using Diagnosis Related Groups (DRGs), which shifted the basis of payment
from one which paid marginal cost approximately in full to one which paid a
lump sum per type of admission, which meant the marginal revenue for an
additional day was zero.
13
Not surprisingly, length of stay fell, but in a few years
bottomed out (Schwartz and Mendelsohn, 1991). Whether there will be an
effect on the steady-state rate of growth in spending remains to be seen. That
will depend upon the degree to which Medicare accounts for new technology in
changing its lump sum over time. Uncertainty about such reimbursement,
11
A much smaller decline in admissions was seen among the elderly, consistent with this story. The
hospital portion of Medicare (Part A) has a deductible that rose in real terms; however, the majority
of Medicare beneficiaries have so-called Medigap or supplementary insurance that typically pays
the deductible. Thus, most Medicare beneficiaries face no effective deductible, and the increases in
the nominal deductible are transformed into increases in premiums.
12
The increased use of caps on out-of-pocket expenditure and increased deductibles left the
out-of-pocket share constant, as noted above. The argument in the text about the marginal price
assumes that price is foreseen. Typically, this will hold if the initial deductible will be small enough
that with high probability it will be exceeded during the hospitalization. There may be some
uncertainty about whether a cap on out-of-pocket payments will be exceeded, however, mitigating
the force of this argument.
13
Marginal revenue is positive for outlier patients, but they are less than 5 percent of cases.
Medicare accounts for just over a quarter of hospital revenues.
Medical Care Costs: How Much Welfare Loss? 15
however, may well decrease spending on the development of new technology
by increasing risk.
The Willingness to Pay for New Technology
Clearly medical expenditures cannot grow forever at a rate 2 to 3 percent-
age points greater than the growth in the economy. Thus, if market forces do
not act to slow the growth, at some point public sector interventions, such as
global budget ceilings, may be desirable. Indeed, legislation contemplating such
ceilings has been introduced. The key issue in appraising such ceilings from a
social welfare point of view is whether consumers are willing to pay for the new
capabilities available to them.
As already noted, the dominant view of health insurance in the economics
literature, at least in the American literature, is that "too much" health insur-
ance leads consumers to demand "too much" medical care at each point in
time,
which is reasonably well established, as well as "too much" technological
change, which is less well established. In two interesting theoretical papers,
Goddeeris (1984a, 1984b) derives the conditions under which a costly techno-
logical change that increased capabilities would increase or decrease welfare. In
the case in which income elasticity for change exceeds zero, surely the interest-
ing case, an innovation that would not be bought without insurance may
nonetheless increase welfare. Whether it does so depends on the marginal
utility of income in the state to which the innovation applies and the number of
low-benefit users induced by the insurance coverage. Although this tells us that
the simple intuition that any subsidy represents a welfare loss does not neces-
sarily apply, it leaves us with the empirical question of how much welfare loss
there might be from the actual rate of change.
Direct evidence is scarce, but some of the evidence cited above suggests the
welfare loss from the additional spending on technology may be less than some
of the rhetoric surrounding the need for cost containment would imply.
First, the real rate of increase in costs is similar across countries, some of
which make centralized decisions about how much to spend on medical care
and all of which have financing institutions that differ from those of the United
States.
14
Countries like Canada and the United Kingdom make explicit bud-
getary decisions about the size of the medical care sector each year and
explicitly trade off medical care against other public sector goods and private
consumption.
14
The U.S. has a large positive residual if current cross sectional health expenditures are regressed
on income (Schieber and Poullier, 1989). A similar regression on earlier data, however, shows the
U.S.
with a much smaller residual (Newhouse, 1977). Thus, relative to GNP growth, U.S. health
expenditure has grown faster than other countries in the 1980s. There is no obvious change in U.S.
financing arrangements in the 1980s that would explain this, however.
16 Journal of Economic Perspectives
Second, the rate of increase in HMO costs in the United States has been
similar to the overall medical care sector. If there were a public preference for
less and cheaper technology, HMOs might have offered it. However, they
apparently do not believe that demand exists for cheaper medical care using a
lower level of technology.
Drawing the inference that consumers are willing to pay for new technol-
ogy from the fact that HMOs have had similar cost increases is open to
argument on several grounds. Some consumers do not face the price difference
between the HMO and traditional reimbursement insurance, because the
employer subsidizes the difference. Others pay the difference with before-tax
dollars. Moreover, offering a markedly outdated technology would open an
HMO to malpractice complaints, since the legal standard used to establish
negligence is the conventional standard of care. Nonetheless, if many con-
sumers felt that new technology wasn't worth the price, it seems odd that we do
not observe some firms trying to enter and offer at least some aspects of 1960s
medicine at 1960s prices.
A Note on The Terminally Ill
One frequently mentioned cause of rising medical expenditure in popular
writing is the treatment of the terminally ill. There is a grain of truth in the
popular perception; among those over 67 years of age, 6 percent died in 1978,
but that group accounted for 28 percent of the expenditure over a two-year
period (Lubitz and Prihoda, 1984).
But taken as a whole, the data offer little support for the notion that society
is wasting an ever-larger share of resources in a fruitless attempt to save those
who are about to die. Spending in the last year of life has not contributed
disproportionately to the
increase
in medical care costs. The share of spending
by those in their last year of life was stable between 1967 and 1979 (Lubitz and
Prihoda, 1984). Moreover, only 6 percent of those who died in 1978 had more
than $15,000 of Medicare expenses, which does not fit the popular image of
heroic efforts being exerted on many terminally ill patients. Put another way,
most dying patients are sick individuals who will receive some sort of care; it
does not appear that a great many of them are receiving enormous amounts of
care.
Furthermore, one usually does not know twelve months before the date of
death that this is one's last year of life. In the late 1970s, physicians were asked
to predict short-term survival probabilities of patients being admitted to an
intensive care unit (Detsky, Striker, Mulley et al., 1981). Those conducting the
study then looked at spending as a function of predicted survival probabilities.
They found that among survivors, expenditure correlated negatively with
predicted survival probabilities (for example, expenditure when the predicted
probability of survival was 25 percent was approximately three times as large as
Joseph P. Newhouse 17
when the predicted probability was 75 percent). Among those who died,
however, expenditure correlated positively with survival probability (for exam-
ple,
expenditure was nearly twice as large when the predicted probability of
survival was 75 percent as when it was 25 percent). Thus, spending was largest
for patients whom physicians expected to live but did not, or for whom
physicians expected to die but did not. These findings are consistent with
rational sequential choice under uncertainty or Bayesian learning; in other
words, many of the 6 percent who spent more than $15,000 may well have
been expected to live.
While the terminally ill undoubtedly absorb some spending of little or even
negative value, it does not appear that such spending is a disproportionate
cause of the increase in medical care costs.
The Uninsured
I now turn from the level of costs to another major health policy issue of
the day, the lack of universal health insurance coverage. About 15 percent of
the U.S. population has no health insurance. The conventional economics view
of this issue is that it is partly distributional; partly an issue arising out of the
connection in the United States between health insurance and one's employ-
ment (for example, those in their 20's seeking their first job have dispropor-
tionately high numbers of uninsured); and partially a market failure resulting
from asymmetric information. As is well known, if the insured knows more
about his or her expected expense than the insurer, there may be no competi-
tive equilibrium, because of adverse selection (Rothschild and Stiglitz, 1976).
Indeed, Akerlof's (1970) classic paper on asymmetric information views the
Medicare program as a public sector remedy for insurance market failure
among the elderly (because they lack the labor-market tie to employer-based
insurance), and Diamond (1992) has recently proposed breaking the connec-
tion between employment and health insurance to remedy selection problems
among the non-elderly.
A program that provided full insurance for the uninsured, according to the
best estimates available, would cause their demand to double (Keeler and
Rolph, 1988). The overall increase in demand for medical care would then be
determined by whether the uninsured are above or below average in their rate
of spending.
The uninsured are a heterogeneous risk group; some of them are individu-
als with a chronic illness who do not have a labor force connection and cannot
purchase individual insurance, or at least insurance that covers their condition.
This group is presumably of above-average risk. Others are people in their 20s
looking for their first job who have chosen not to purchase individual insurance
in part because they believe they are healthy. This group is presumably of
below-average risk. On balance the uninsured as a group probably are
18 Journal of Economic Perspectives
somewhat above average
in
their risk.
On the
other hand, partial rather than
full insurance coverage
is a
more likely scenario
in any
intervention
to
cover
the
uninsured.
Taken
all
together, covering
the
uninsured (exclusive
of
chronic long-term
care) would probably increase demand
for
medical care
by
less than
10
percent,
or
the
equivalent
of
roughly
two
years
of
real growth
in
health expenditure
at
historical rates. Thus, universal insurance, while adding nonnegligible costs,
would
not
wreak havoc
on the
nation's medical care economy.
The method
of
financing
any
increase
in
medical care costs, whether those
costs stem from future technological change
or
from covering
the
uninsured,