Chronic Illness in a Complex Health Economy:
The Perils and Promises of Downstream and Upstream Reforms
Jack Homer, Gary Hirsch, and Bobby Milstein
Abstract
Chronic illness is the largest cause of death and source of health care costs in developed
countries and has become a significant problem in developing countries as well. This
paper begins with a review of past work in System Dynamics concerning populations
with chronic illness. It then presents a generic model of illness in a population and its
treatment and prevention, applied to the U.S. population.. The model explains the rising
prevalence of illness as well as responses to it, responses which include the treatment of
complications as well as disease management activities designed to reduce the occurrence
of future complications. The model shows how progress in complications treatment and
disease management has slowed since 1980 in the U.S., largely due to a behavioral tug-
of-war between health care payers and providers that has resulted in price inflation and an
unstable climate for health care investments. The model is also used to demonstrate the
impact of moving “upstream” by managing known risk factors to prevent illness onset,
and moving even further upstream by addressing adverse behaviors and living conditions
linked to the development of these risk factors in the first place.
Introduction
Chronic illness is a major health challenge facing all countries. It is the largest cause of
death and source of health care costs in developed countries and has become a significant
and growing problem in developing countries as well (Mackenbach 1994, Olshansky and
Wild et al 2004, Mathers and Loncar 2006). In the U.S., the Centers for Disease Control
and Prevention (CDC) estimates that chronic illness is responsible for 70% of all deaths
and 75% of all health care costs (CDC/NCCDPHP 2007a). The aging of the U.S.
population and increases in risk factors such as obesity suggest that chronic illnesses will
be an even greater problem in future years. Already, according to the CDC, an estimated
32% of U.S. adults are obese (CDC/NCCDPHP 2007b). Moreover, the fraction of
children who are overweight has tripled to 16% during the last 20 years. The good news
is that mortality rates from chronic illness have fallen significantly since 1970, dropping
by about half for heart disease and stroke for example (NIH/NHLBI 2004). Even this
good news must be tempered, however, since it means there are many more people living
with chronic illness and its associated disabilities and health care costs.
Worldwide, the trends are even more stark. The World Health Organization (WHO)
reports that 80% of deaths from chronic illness occur in lower- and middle-income
countries (WHO 2005). In many developing countries, deaths from chronic illnesses now
outstrip mortality from traditional health concerns such as injuries and communicable
diseases (Yach et al 2004). And rates of chronic illness are rising in the developing
countries, creating an additional burden of disease on top of high rates of acute illness.
The worldwide prevalence of diabetes, for example is projected to rise from 171 million
(2.8%) in 2000 to 366 million (6.5%) in 2030, with over 80% of the projected cases in
2030 occurring in the developing world (Wild et al 2004). Future economic development
is expected to bring with it increased risk of morbidity and mortality tied to chronic
illness and driven by growth in obesity, tobacco use, and other risk factors.
In most nations, health care systems are organized in a way that makes them hard-pressed
to respond to chronic illness. The shortcomings of health systems in dealing with chronic
illness include a failure to empower patients and involve them in their own care, a lack of
linkages between the health care system and other community agencies that should be
involved, misaligned incentives for providers, and a failure to invest in prevention (WHO
2002). In the U.S., the Institute of Medicine has detailed changes needed in the health
care system to effectively cope with the increasing burden of chronic illness, including
consistent provision of evidence-based care, reorganization of clinical office practices to
provide for longer visits needed for patient education and follow up, attention to the
needs of patients when seeking lifestyle and other behavioral change, and implementation
of supportive information systems (IOM 2001).
This paper begins with a review of past work in System Dynamics (SD) concerning
populations with chronic illness. It then presents a generic model of illness in a
population and its treatment and prevention, applied to the U.S. population. This model
encompasses not only chronic illness, but all illness and injury, primarily because of data
limitations discussed below. The distinction between chronic and acute conditions is a
somewhat arbitrary one, anyway. Some chronic illnesses can nowadays be cured quickly
once they are discovered, and conversely, some acute infections or injuries, if not treated
quickly, can become chronic problems. The combining of all afflictions into a single
model requires only that the rates of death and cure reflect the entire continuous
distribution of illness, from the very short-lived to the very long-lived, and from the
easily cured to the incurable. Although our model covers all manner of illness and injury,
our focus is on those chronic illnesses which are long-lived and incurable, which are
responsible for the great majority of health impairment in the U.S.
Our model explains the rising prevalence of illness as well as responses to it, responses
which include the treatment of complications as well as disease management activities
designed to slow the progression of illness and reduce the occurrence of future
complications. The model shows how progress in complications treatment and disease
management has slowed since 1980 in the U.S., largely due to a behavioral tug-of-war
between health care payers and providers that has resulted in price inflation and an
unstable climate for health care investments. The model is also used to demonstrate the
impact of moving “upstream” by managing known risk factors to prevent illness onset,
and moving even further upstream by addressing adverse behaviors and living conditions
linked to the development of these risk factors in the first place.
Applications of System Dynamics to Chronic Illness
A number of applications of SD to chronic illness extending over three decades provide a
foundation for the concepts discussed in this paper. Dental care and oral health was a
focus of early work. The most expansive of these studies (Hirsch et al 1975) explored
feedback relationships among the supply of personnel and the availability of care, the
distinction between preventive and symptomatic care, the oral health status of a
population and prevalence of dental disease, and the workload of dental practices. This
study also analyzed the impacts of various dental manpower policies on oral health
outcome measures including prevalences of decayed, missing, and filled teeth.
Several SD modeling efforts have focused on cardiovascular disease. A model
developed for the State of Indiana Health Planning Agency (Hirsch and Myers 1975)
projected the prevalence of heart disease and stroke in the state and evaluated the
potential impact of different programs for reducing the costs and mortality due to these
diseases. The model represented multiple stages through which people move as they
develop cardiovascular disease from predisposing conditions such as hypertension to
undetected and nonacute illness, acute incidents such as heart attacks and strokes, and
rehabilitation and recovery after such attacks. Simulations with the model illustrated the
value of comprehensive programs that combine preventive interventions such as
hypertension screening and treatment with improved acute care.
Another model of cardiovascular disease (Luginbuhl et al 1981) used a similar structure
to examine the impact of investing more resources in prevention and rehabilitation rather
than more elaborate technologies for treating acute myocardial infarction. The model
demonstrated how prevention and rehabilitation could lower the costs of heart disease in
the U.S. more effectively than new technologies that only marginally extend the lives of
people who are in the later stages of the disease.
Diabetes is another area in which SD modeling has been used to study chronic illness in
populations. A model developed for a community coalition in Whatcom County in the
state of Washington (Homer, Hirsch, et al 2004) portrayed patients flowing through
several stages as they moved from being at-risk for diabetes into diabetes and its
complications and moved from having their blood sugar levels not under control to under
control. The model demonstrated how the right combination of interventions for
prevention and treatment could reduce the burden of diabetes in terms of both mortality
and cost. A similar population flow model of congestive heart failure—which, like
diabetes, is another chronic illness producing high burden in the U.S.—was developed for
Whatcom County and used for a similar analysis of interventions.
Another SD model of population flows in diabetes (Jones et al 2006) was conducted for
the CDC and developed with experts at the federal, state, and local levels in the U.S.
This model is similar to the Whatcom County diabetes model in many ways, but enables
a closer look at primary prevention by delineating the condition of moderately elevated
blood sugar known as prediabetes and by portraying the significant influence of obesity
(the leading modifiable risk factor for diabetes) on the onset rates for prediabetes and
diabetes.
Other prominent SD models exploring the epidemiology of particular chronic conditions
addressed obesity (Homer et al 2006) and smoking (Tengs 2001).
Some SD modeling has considered chronic illnesses more generally, rather than focusing
on a specific disease. A Health Care Microworld developed by the New England Health
Care Assembly and Innovation Associates (Hirsch and Immediato 1999) portrays a
population at different ages as they develop and move through increasingly severe stages
of chronic illness. Users of the Microworld can employ a variety of medical and non-
medical interventions to influence these population flows, including interventions that
can mitigate social, behavioral, and environmental risk factors for chronic illness.
The common feature of many of these earlier efforts is the focus on a population
developing a specific illness and then moving through one or more stages of increasing
severity and complications. Movement between these stages occurs at rates that depend
on behavioral and environmental factors as well as demographic characteristics. The
models allow for multiple points of intervention, both downstream after the disease
process has ensued and upstream at points when disease incidence can still be prevented
or the well-being of people better protected. A common lesson is the value of balanced
strategies that include preventive programs as well as care and treatment to produce the
most net benefit in both the short term and the long term. In the next section, a model is
presented that, aggregating across all illnesses, demonstrates the potential impacts of
attempting to improve downstream care or upstream prevention and describes the
economic mechanisms for such interventions.
A National-Level Model of Downstream Care and Upstream Prevention
Model Scope and Historical Evidence
The shortcomings of the U.S. and other health care systems in dealing more effectively
with chronic illness are systemic and not confined to particular localities or particular
illnesses (IOM 2001). They arise from the interactions of multiple stakeholders,
including patients, providers, employers, third-party payers, makers of products, and
regulatory and monitoring bodies as well as groups of ordinary citizens. Some of these
actors are very influential, and their decisions can affect the health of an entire nation.
Accordingly, we have chosen to develop a model at the national level that aggregates
across all illnesses to explore questions related to the evolution of downstream care and
the potential benefits and costs of greater upstream effort. This model is based on data
specific to the U.S., but its structure should be applicable to other countries. The past
applications of SD to chronic illness have served as useful background for this work, as
have broader SD and non-SD studies that have considered the dynamics and economics
of population health without a particular focus on chronic illness. The SD studies of this
sort that have contributed to our thinking include (1) a simulation model of U.S. health
care spending and finance (Ratanawijitrasin 1993), (2) a simulation model of community
health and the “syndemic” confluence of multiple interacting afflictions (Homer and
Milstein 2002, 2004), (3) a conceptual framework for thinking about U.S. health care
reform (Hirsch, Homer et al. 2005), and (4) a conceptual framework for thinking about
upstream-downstream dynamics (Figure 3 in Homer and Hirsch 2006). The influential
non-SD studies of health include books by Starr (1982) and Heirich (1999) and articles
by Weisbrod (1999) and Cutler et al. (2006).
The SD works considering the health system broadly have contributed useful ideas and
hypotheses. In our present work, we have looked more closely at historical data and
sought to develop a model capable of reproducing key elements of that history so that we
may better understand its underlying causes. While such empirical grounding does not
guarantee that the model is adequate and useful for exploring the future, it is an important
step toward that end (Homer 1996).
As we have gained familiarity with the historical data, we have come to focus our
modeling effort on a perplexing question: Why, with the tremendous growth in health
spending since 1960, is the health of Americans not better than it is? More specifically,
why has the U.S. health care system, for all its size and capability, not managed to subdue
chronic illness more effectively?
A key source of historical data has been the National Health Expenditure Accounts
(NHEA) (CMS 2007), which measure annual health spending in the U.S. by category.
From 1960 to 2004, total health spending grew (in per capita, constant dollar terms) by a
factor of eight, and as a fraction of gross domestic product (GDP) tripled from 5.2% to
16.0%. Note that we are no longer speaking of chronic illness alone: the NHEA data
cover all health spending and do not distinguish between expenditures for chronic illness
and those for acute illness and injury. Although estimates of national spending exist for
some individual chronic illnesses such as diabetes (ADA 2003), these are generally only
on a one-time snapshot basis, and no comprehensive running audit of overall chronic
illness spending is performed. Given this situation, and not wishing to abandon our
desire to be empirically grounded, we have decided to expand the purview of the model
to include all illness and injury, and not only chronic illness.
Total health spending grew rapidly from 1960 to 1990, slowed during the 1990s, then
resumed more rapid growth in 2000. A consistent 82-85% of total health spending has
been for what is known as personal health care (or what one might call health care
consumption), which comprises hospital care (30% of health spending in 2004),
nonhospital services (37%), drugs and health-related products and equipment purchased
for individual use (13%), and miscellaneous personal health care (3%). Components of
health spending in the NHEA other than personal health care include administration,
public health, research, and capital investments. Rising costs for outpatient care have
been responsible for much of the growth in health spending in the U.S. since 1980.
The recent historical record suggests the health of Americans has not improved as much
as one might have expected from the dramatic growth in health care spending. We define
illness or disease as a moderately or severely symptomatic biological or psychological
condition—i.e., one associated with some reduction in perceived health-related quality of
life. (A person with an asymptomatic or only mildly symptomatic condition is
considered to be at risk for disease. Although not yet considered to have full-fledged
disease, that person may be eligible for management or treatment of the risk condition.)
Two of the CDC’s large national annual health surveys, the National Health Interview
Survey (NHIS) (CDC/NHIS 2007) and the Behavioral Risk Factor Surveillance System
(BRFSS) (CDC/BRFSS 2007), report the fractions of individuals describing their own
health status as excellent, very good, good, fair, or poor. Research shows that these self-
reported health metrics have desirable statistical properties and are predictive of adverse
health events (Dominick et al. 2002). The NHIS also publishes the self-reported
prevalence of common chronic conditions. We have examined the reported results of
other national surveys as well (Thorpe et al. 2005, Hoffman et al. 1996).
After considering all of these data sources, we have concluded that the NHIS sum of the
poor, fair, and good health status categories (that is, people not reporting their health as
excellent or very good) is the best indicator of the prevalence of illness as we have
defined it, with a continuous span of reporting from 1982 to the present. Throughout this
time period, this sum has remained within the relatively narrow range of 31% to 35% of
the population, with some movement downward through 1990, upward through 1993,
downward through 1998, then upward through 2004. Because the periods of downward
movement are not consistent with some of the other measures described above, we are
reluctant to emphasize the NHIS fluctuations before 1998. But the upward movement
from 31% in 1998 to 33% in 2004 clearly is consistent with the other NHIS and BRFSS
measures. We are thus confident in saying two things about the prevalence of illness
since 1982: (1) it has not varied by much and certainly has not declined significantly, if at
all; and (2) it has increased somewhat since the late 1990s.
To address the question of why the U.S. has not been more successful in preventing and
controlling chronic illness, we have constructed a simulation model that, although still a
preliminary theory, can faithfully reproduce observed patterns of change in disease
prevalence and mortality, and that can also reproduce the histories of the model’s primary
explanatory variables. The full model contains about 200 equations, including 9 stocks
and delay functions, 50 constants, and 11 exogenous time series. (The Vensim model is
available in the online supplement, or upon request from the authors.) Some of the
exogenous time series ensure a closer model fit to history, while others represent
potential policy levers. The exogenous time series do not affect the general findings
discussed below; these findings are entirely determined by the model’s feedback
structure.
Conceptually, the model’s hypothesized causal structure can be considered in three parts:
(1) a population stock and flow structure; (2) feedback structure that explains the past and
especially the growth of downstream care and spending; and (3) additional structure that
can help explore the benefits and costs of upstream efforts to improve health.
Population Stocks and Flows (Figure 1)
Figure | depicts all members of the population as being in one of three stocks: not at risk,
at risk, or with disease. The population is increased by a net inflow rate, corresponding
to births plus net immigration, and assumed for the sake of simplicity to flow entirely into
the stock of population not at risk. The population is decreased by deaths, which are of
two types: (1) deaths following disease, affecting only the stock of population with
disease); and (2) deaths from injury or violence, assumed to affect equally all three
population stocks in the model. In 2003, the 1-year odds of death from injury or violence
were | in 1,743 and deaths from these causes accounted for only about 7% of all deaths
(NSC 2007, CDC/NVSR 2006).
Disease prevalence is the fraction of the population with disease, while risk prevalence is
the fraction of the population at risk. Disease is defined above. Risk refers to physical or
psychological conditions or individual behaviors that may lead to disease. In particular,
we have used as a proxy measure for risk prevalence the fraction of the adult population
with one or more of the cardiovascular risk factors of hypertension, high cholesterol,
hyperglycemia, obesity, and smoking. BRFSS data from self-reports indicate that this
measure of risk prevalence grew continuously during the period 1991-1999, rising from
58% to 62% (Greenlund et al. 2004).
Flows between the stocks, as well as disease-related deaths, may be affected by certain
actions and factors we will discuss in the remainder of this paper. The disease-related
death rate is affected by the effectiveness of urgent care for disease complications,
generally involving hospitalization. The frequency of complications, in turn, is affected
by the effectiveness of disease management. In some cases, effective disease
management may increase the likelihood of disease cure or recovery; this is certainly true
for many acute infectious diseases and can also be true for chronic diseases, as in the
cases of organ transplantation or cancer chemotherapy. Effective risk management can
reduce the flow of people from risk to disease, and may also in some cases allow people
to return to a condition of being no longer at risk. Such management may include
changes in nutrition or physical activity, stress management, or the use of medications.
Flows of risk onset and risk reduction are affected by adverse behaviors and living
conditions. Adverse behaviors may include poor diet, lack of physical activity, or
substance abuse. Adverse living conditions can encompass many factors, including
crime, lack of access to healthy foods, inadequate regulation of smoking, weak social
networks, substandard housing, poverty, or poor educational opportunities. In calibrating
the model, we have found that the rise in risk prevalence for 1991-1999 described above
can be explained by assuming that the onset of risk due to adverse behaviors and living
conditions increased by 30% from 1980 to 1995, and by another 5% through 2005. The
timing and shape of this increase corresponds well to the apparent historical pattern of
growth in net caloric intake that has driven the rise in obesity in the U.S. since the late
1970s (Homer et al. 2006).
Downstream Loops (Figure 2)
Figure 2 presents a theory of the growth of downstream care and spending. This growth
is affected by changes in disease prevalence, as well as by changes in the extent of care
(disease management and urgent care), and in health care prices.
What are the drivers of extent of care? For the purposes of our model, we have reduced a
large literature on health care quality (see IOM 2001), of which extent of care is a part,
down to just two factors: the abundance of health care assets, and insurance coverage.
By health care assets we mean the structures and fixed equipment used directly for health
care or for the production of health care products, as well as the human capital of
personnel involved. A greater abundance of assets nationwide means that a larger
number of people have access to a broad array of medical services, but beyond a certain
point some of that greater abundance represents duplication, and as a result one reaches a
point of diminishing returns to extent of care.
By insurance coverage we mean the fraction of the population with some form of health
care insurance, either with a private insurer or through a government plan. (Government
plans are available in the U.S. for the lower income, the elderly, the disabled, and for
military personnel and war veterans.) The uninsured are less likely than the insured to
receive health care services. The effect of insurance on extent of care is modeled as
being relatively strong in the case of disease management services, for which the vast
majority of providers require payment (something most of the uninsured cannot afford);
while the effect is weaker in the case of urgent-care services, reflecting the fact that
hospitals in the U.S. are required to provide emergency-department access even to
patients unable to pay.
Health Care Assets:
The model includes two separate stocks of health care assets which differ in terms of
their uses: a stock used for disease or risk management, and a stock used for urgent
(complications) care. These two stocks have likely grown at different rates at different
times over the years. Distinguishing these two stocks and their different growth rates in
the model has helped us to explain the evolution of health care spending evident in
NHEA data, from more rapid growth in urgent care in the 1960s and 1970s, focused on
hospital-based life-saving interventions, to more rapid growth in disease and risk
management since about 1983, focused on the development and use of diagnostic
equipment and pharmaceuticals.
To calibrate the asset sector of the model, we have looked primarily to NHEA data on
investments in structures and equipment (S&E). (We have found no data set on human
capital in health care that is complete and can be harmonized with the data on structures
and equipment.) In particular, we have estimated (via spreadsheet calculations) the net
value of health care assets by accumulating the S&E investments over time,
decrementing for obsolescence or depreciation at an assumed rate of 5% per year, and
initializing in 1960 at a level that permits smooth early growth of the estimate. The
resulting estimate grows at an average rate of 4.1%/year during 1960-1980, and
3.2%/year during 1980-2004. This growth in assets is consistently less than that of
personal health care spending (consumption), which grew 5.4%/year during 1960-1980
and 4.3%/year during 1980-2004. We hypothesize that this difference reflects a decline
over time in the fraction of health care revenues reinvested in assets. In fact, we have
found it is possible to reproduce the estimated time series for health care assets by
assuming that the revenue reinvestment rate declined from 13-14% in the 1960s to 10%
in 1980 to 6% in 2004.
Why should the revenue reinvestment rate have declined in this way? We suggest that
the cutback in investment has been the response by potential investors to various forms of
cost control, including the restriction of insurance reimbursements, which affect the
providers of health care goods and services.’ With increasing controls and restrictions,
these potential investors face greater risk and uncertainty about the future return on their
investments, and the result is a greater reluctance to build a new hospital wing, or to
purchase an expensive new piece of equipment, or even, at an individual level, to devote
a decade or more of one’s life to the hardship of medical education and training.
Taking one step back, why is it that cost controls started to take hold in the 1970s and not
earlier? Several authors (e.g., Starr 1982, Eckholm 1993, Heirich 1999) have described
how economic power, starting in the 1970s, shifted from providers of medical care, who
had been allowed to act freely for many decades, to employers and public agencies
desiring to rein in costs. As Paul Starr (1982) puts it, “Until the 1970s. ..practitioners,
hospitals, researchers, and medical schools enjoyed a broad grant of authority to run their
own affairs. In the 1970s the mandate ran out.” Max Heirich (1993) describes this shift
as a reaction to the growth in health care costs relative to the rest of the economy
beginning in the 1960s: “Where for decades [the costs of American health care] had
consumed between 3.5 and 4.5 percent of GNP...by 1960 its share of the GNP had
increased to 5.3 percent...and health care’s share of GNP increased to 7.3 percent in
1970...The American health-care system’s non-equilibrium growth in costs now affected
the rest of the economy.”
Health Insurance Coverage:
While some employers have reacted to high health care costs by selecting less generous,
more cost-restrictive insurance plans for their employees, others have taken the more
drastic action of not providing coverage at all to many of their workers. Surveys of
insurance coverage taken annually since 1987 (Census 2007a) show that the fraction of
the U.S. population covered to some degree by private (employer-provided or self-
purchased) insurance fell from 75.5% in the late 1980s to 70% through most of the
1990s, rose briefly to 72% during 1999-2000, then declined again to 67.7% by 2005.
This decline in private coverage is a serious matter affecting the ability of tens of millions
of Americans to gain access to regular, good-quality health care. However, there is
another dimension to the insurance story, and that is the growth of government-provided
insurance. This growth started with the passage in 1965 of the federal Medicare and
Medicaid programs to provide coverage for elderly and lower income people,
respectively. The Medicaid program in particular has grown over the years in terms of
the fraction of the population it covers, from about 8.4% in 1987 to 12% in the early
1990s, declining to 10% in the late 1990s, and then rebounding to 13.0% by 2005. Thus,
the Medicaid curve has for nearly 20 years moved consistently in a direction opposite to
the curve for private insurance: a decline of 7.8 percentage points in private coverage has
been countered by an increase of 4.6 percentage points in Medicaid coverage. As a
result, the fraction of the population with any insurance coverage, private or public, has
fallen by only 3 percentage points, from 87.1% in 1987 to 84.1% in 2005. Clearly, many
of the people who have lost coverage from their employer or as a result of changing jobs,
primarily wage-earners in lower paid positions, have been able to switch over to
Medicaid as a fallback.”
Let us review the story of the health care system’s evolution told thus far by walking
through the hypothesized feedback loops in Figure 2:
= Loop R1 shows how the funds generated by health care lead to more investment
i ets, and how the application of these new ‘ts in the form of more
extensive care can generate even more funds to support further growth. Even in
today’s more restrictive climate, this loop remains central to the story of progress
in health care.
= Loops B1 and R2 show how more extensive care has effects on health and
longevity that can moderate or reinforce Loop R1. Loop B1 indicates that
increased disease management can prevent costly complications and thereby
reduce spending and the need for investment in new assets for urgent care. Loop
R2 indicates, however, that insofar as more extensive care prolongs life for people
with disease, it tends to increase disease prevalence and thereby increase
spending and investment in health care assets.
= Loop B2 shows how rising personal health care spending as a fraction of GDP
triggers a backlash from employers and other payers, resulting in a more
restrictive reimbursement climate that can suppress the rate of investment in new
assets and thereby slow the growth in health care costs, although at the same time
slowing further growth in the extent of care.
= Loop B3 shows how the denial of insurance coverage by some employers in
reaction to high health care costs appears to be another route for slowing the
growth an those costs, although, like Loop B2, it also slows growth in the extent
of care.”
Taken together, one may view these loops as the story of a health care system that favors
growth and investment until the resulting costs get to a point where further increases are
perceived to be no longer worth the expected incremental improvements in health and
productivity. That does not by itself sound like a story of dysfunction but rather one of
progress followed by goal-seeking behavior. There is a potential for dysfunction in Loop
B3, where a reduction in insurance coverage can drive up the unreimbursed costs of
hospitals (resulting in a burden on the general public), and also create a situation of health
inequity that separates the uninsured poor from the rest of society. But, although the
insurance gap is certainly a matter of concern, that gap has been with us for decades, and
its growth by 3 percentage points since 1987 is not by itself alarming. Because of this
small magnitude of change, declining insurance coverage is unlikely to contribute much
me |
toward answering our question of how it is that health care spending can keep growing
without doing much to improve health for the majority of the population.
Health Care Prices:
To find a more compelling causal mechanism behind this sort of system failure, we must
go one step further and consider the dynamics of health care prices. Medical care is one
of eight major groups in the Consumer Price Index (CPI) computed by the U.S. Bureau of
Labor Statistics (BLS), measuring retail price changes over time “for a constant quality,
constant quantity market basket of goods and services.” (BLS 2007). The medical care
CPI combines four major components, with approximate importance weights for 2005 as
follows: professional services (2.8), hospital services (1.6), drugs and other personal use
products (1.5), and health insurance (0.4). The medical care CPI has grown more rapidly
than the general CPI for the overall economy, especially since 1980. For 1960-1980,
inflation in medical care prices averaged 6.2% compared with general inflation of 5.3%,
while for 1980-2004, inflation in medical care prices averaged 6.1% versus general
inflation of 3.5%. Consequently, a fixed market basket of medical care goods and
services costing $100 in 1960 had risen to $1,391 in 2004; while a market basket for the
general economy costing $100 in 1960 had risen to $638 in 2004.
Why has health care inflation exceeded that of the general economy? We have
considered various possible explanations for why costs should have gone up so rapidly,
particularly since 1980, for a given quality of care. These include increasing costs for
drug development; more gadgetry in medical technology; the increased practice of
“defensive medicine” by providers to avoid lawsuits alleging malpractice; the increase in
medical malpractice insurance premiums; the shift of many procedures from inpatient
settings to outpatient settings where prices may be less tightly regulated; and the use by
providers of various methods to maintain their incomes in the face of greater restrictions
on reimbursement. Although all of these phenomena have contributed to health care
inflation, not all have contributed with sufficient magnitude or with the timing necessary
to explain the historical pattern. One phenomenon that does appear to have such
explanatory power, and which we have centered on for the purposes of this study is the
last one listed above, described in Figure 2 as “provider adaptation”, or elsewhere as “the
target income hypothesis” (Ratanawijitrasin 1993, p. 77) or “the behavioral response”
(Peter Passell in Eckholm 1993, p. 285).
A variety of studies since the late 1970s provide strong support for the idea that, in
response to cost containment efforts, providers may “increase fees, prescribe more
services, prescribe more complex services (or simply bill for them), order more follow-up
visits, or do a combination of these...” (Ratanawijitrasin 1993) Specific billing practices
that can circumvent cost containment efforts include “upcoding” (billing with procedure
codes that receive higher reimbursement rates) and “unbundling” (billing a single
procedure in multiple parts to achieve a higher total) (Eckholm 1993). Many tests and
procedures are performed that contribute little or no diagnostic or therapeutic value,
thereby inflating the cost per quality of care delivered. Writing in the New York Times in
April of 1989, the former Secretary of Health, Education, and Welfare, Joseph Califano
Jr., “claimed that Americans would spend about $155 billion in 1989 for tests and
=
treatments that would have little or no impact on the patients involved.” (Heirich 1999; p.
97) If correct, that unnecessary and inflationary expense would have represented 29% of
all personal health care spending in that year.
Increased pressure on provider incomes comes not only from reduced reimbursements,
but also from the administrative burden of dealing with many different insurance plans.
With the era of cost containment also came greater competition between private insurers
to offer employers acceptable benefits for their employees at the lowest price. One
aspect of this competition is the creation of a broad and ever-changing menu of plans
with different exclusions and different payment percentages for different health services.
With this cacophony of payer fee schedules, the administrative overhead of providers in
the U.S. has grown enormously, threatening to reduce provider incomes. (Woolhandler
et al. 2003 estimates administrative costs as 31% of provider revenue in the U.S.
compared with 16% in Canada.) Providers have thus felt even more need to maintain
their incomes through adaptation, and have consequently driven inflation in health care
prices even further.
With the inclusion of provider adaptation in Figure 2 to explain health care inflation, a
new loop is created: Loop R3. This loop describes the tug-of-war between payers
restricting reimbursement in response to high health care costs, and providers adapting to
these restrictions by effectively raising health care prices in an attempt to maintain their
incomes. This loop has the effect of reducing the efficiency of health care spending and
thus artificially raising the cost of health care to payers. The payers react to the
magnified costs by seeking further restrictions on reimbursement, or by further denying
insurance coverage. The net result is a reduction in health care assets and insurance
coverage (through Loops B2 and B3, respectively), thus dampening growth in the extent
of care. As shown below, this unintended chain of events might have been avoided or at
least moderated had payers and providers not set Loop R3 in motion.
Baseline Simulation (Figure 3) and Alternative Tests of Downstream Behavior (Figure 4)
In Figure 3 we present results from the baseline simulation for several of the model’s key
variables along with historical data. Results from the model are shown from 1960
through 2010." We recognize that a couple of these data series are conceptually
incomplete. In particular, the measure of health care investments does not include human
capital, and the measure of the population at-risk fraction is based only on adults and on
cardiovascular risk factors. Although more complete measures would likely show the
same sorts of trends and have little or no effect on model findings, we would like to
construct more complete data series, if possible, in future iterations of our model.
Having established the model’s ability to do a good job of reproducing historical trends
for a variety of key variables, let us examine how a few of the key feedback loops in
Figure 2—in particular, those depicting the reactions of payers and providers—contribute
to the overall simulated behavior. Shown in Figure 4 are results from the base run
alongside results from alternative simulations for 1960-2010 in which one or more of
=
these feedback loops has been cut. The assumptions and results for the simulations are as
follows:
No coverage down: In this simulation, employers do not react to high health care costs
by denying private insurance coverage; Loop B3 is cut. As a result, the insured fraction
of the population does not decline, as it does in the base run; instead it continues to climb
gradually to reflect the increasing availability of Medicaid coverage to those with lower
incomes. With increased coverage, the extent of care—particularly disease
management—is improved, and the rate of urgent episodes is therefore lower than in the
base run. But this more extensive care costs more than it saves, and thus health care costs
per capita increase relative to the base run. This outcome would seem to suggest that
employers who have denied coverage to their employees have thereby saved money.
Note, however, that the costs in the model do not include the sick days and losses of
productivity that are much more likely to occur when disease is not well managed. This
is why some of the nation’s employers are taking another tack, providing free or low-cost
primary health care in their own offices as a way of improving productivity and catching
health problems before they get more serious and require expensive outside care
(Freudenheim 2007).
No price up: In this simulation, providers do not react to restrictions on reimbursement
by raising their fees for a given quality of service; Loop R3 is cut. As a result, health
care costs per capita grow much less than in the base run. Lower costs mean fewer
revenues available for reinvestment but also less restriction of reimbursements and
coverage. Because reimbursements are more stable, the investment rate does not decline
as much as in the base run, and so, despite the decline in the revenue base, assets per
capita increase no less than in the base run and even a bit more. With lower costs, there
is also much less denial of insurance coverage. Because of the greater insurance
coverage and the slightly greater assets, the extent of care is improved and urgent
episodes per capita are reduced relative to the base run.
This simulation points to the importance of the “dysfunctional” Loop R3, but its results
should not be taken too literally or as a prescription for policy. Legislators seeking to
stabilize health care costs might be tempted to limit the autonomy of providers when it
comes to billing and compensation, requiring that they be paid a fixed amount (as is done
in some managed care organizations), perhaps through a single government payer (as is
done in many countri Some providers in the U.S. might welcome the predictability
and reduced administrative burden such a simplified payment system would bring.
Others, however, are likely to protest such loss of autonomy, especially the many who
expect (and whose adaptive behavior to date has been based on the expectation of) high
incomes in return for their long years of education and training. A national fixed-price
policy might therefore be met by a decline in the supply of providers—an increased rate
of retirement and decreased influx of medical students—leading perhaps to a severe
shortage. In terms of Figure 2, if the adaptive responses of providers were no longer
permitted, we might see a decline in the human capital component of health care assets;
that is, a strengthening of Loop B2. Such a reaction could conceivably cause a fixed-
price policy to do more harm than good, if the reaction were strong enough.
mit @
No reimburse down: In this simulation, employers and other payers do not react to high
health care costs by restricting reimbursements; Loops B2 and R3 are cut. The stable
reimbursement climate encourages more investment in assets as a fraction of revenues
and also defuses the dysfunctional tug-of-war between payers and providers that leads to
price inflation in Loop R3. The cutting of Loop R3 does keep health care costs down (as
in the No price up simulation), but in so doing reduces health care revenues and therefore
initially counteracts the effect of an increased investment fraction on asset formation,
relative to the base run. By the 1980s, however, the stable investment fraction more and
more differentiates this scenario from the base run, and assets thus start to grow faster
than in the base run. The rapid growth in assets per capita drives greater improvements
in extent of care so that urgent episodes decline much more than in the base run. Also,
the lower health care costs relative to the base run (until late in the simulation) mean that
there is less loss of insurance coverage, which improves the extent of care further. Not
until midway through the 2000s does the rapid growth of assets finally increase health
care costs above where they are in the base run, causing erosion in the insured fraction
back to the base run level by the end of the decade.
The No reimburse down simulation underscores the importance of the dysfunctional
payer-provider interaction in Loop R3 and also points to the importance of the impact of
payers on investors in Loop B2. But, as above, the results should not necessarily be
viewed as having direct policy implications. They seem to suggest—perhaps
counterintuitively—that health insurance should be stable and nonrestrictive in its
reimbursements, so as to avoid behavioral backlashes that can trigger health care inflation
and under-investment. However, few policymakers in the U.S. would likely be willing to
mandate that private payers must provide plans of only a certain sort (if they are going to
provide insurance at all), as such a mandate would be seen as interference in a matter of
private choice. Perhaps, then, the mandate could apply only to the government’s own
insurance programs. (Government reimbursement practices are often copied by private
insurers, and so with such an approach one may end up with the desired effect on the
private sector without having to interfere with it.) Even so, many policymakers might
fear that such a mandate would open the door not only to beneficial investments, but also
to indiscriminate and wasteful ones, such as occurred most prominently before the era of
cost containment. Still, it is interesting to consider whether a more generous and stable
approach to reimbursement could not only combat illness better than the current
restrictive approach, but do it more efficiently and perhaps even at lower cost.
The above analysis suggests that there are no easy downstream fixes to the problem of an
under-performing and expensive health care system. It is one thing to understand the
dysfunctional tug-of-war between payers and providers, but quite another to defuse it.
We have addressed the lagging extent of care in our model by looking at the influences of
health care assets and insurance coverage, but we have not explored improvements in the
efficacy and safety of that care. Such improvements can include better information and
decision-support systems, better payment incentives, and better clinical training. Local
implementations of such improvements indicate their promise for reducing the burden of
disease and providing more effective care for the health care dollar (IOM 2001). One
-14-
wonders, though, just how much we can hope to gain from such downstream measures,
when they may appear to payers or providers as an even greater expense to bear (at least
initially) and could therefore end up feeding into the system’s divisiveness and
dysfunction.
Potential Upstream Loops (Figure 5) and Tests of Their Behavior (Figure 6)
Let us turn, then, to the upstream prevention of disease incidence, to see what promise it
may hold for lessening our dependence on a costly and inefficient system of downstream
care (Fries et al 1998, McKinlay 1979, McKinlay and Marceau 2000). Illustrated in
Figure 5 are two broad categories of such efforts: Risk management for people already at
risk, and health protection for the population at large. The literature identifies significant
opportunities for medically oriented risk management for a variety of diseases, through
improved nutrition and exercise, smoking cessation, and the appropriate use of drugs
(Eyre et al 2004, Hajjar et al 2006, Leonhardt 2007). For example, the fraction of people
with hypertension whose condition is considered under control (based on data on blood-
pressure measurement from the National Health and Nutrition Examination Survey) stood
at 29% for 1999-2002, up only a few percentage points from 25% for 1988-1991 (Hajjar
et al 2006).
The literature also describes opportunities for socially oriented health protection, which
may include efforts to change adverse behaviors and mitigate unhealthy conditions in
homes, schools, workplaces, and neighborhoods and to alter macroeconomic forces and
the media so that they are more health promoting (Northridge et al 2003, Gerberding
2005, Yach et al 2005, Homer, Milstein, et al 2006, Simon 2006, CDC 1999, CDC 2006,
IOM 2002, Smedley and Syme 2000, Wilkinson and Marmot 2003, Evans et al 1994,
Hanna and Coussens 2001). Note that unlike downstream interventions, health protection
efforts rely on the actions of individuals and organizations most of whom are not health
care professionals.
What can the data tell us about the history of upstream spending? Much upstream work
involves population-based public health efforts emphasizing health promotion and
disease prevention. Public health spending has grown as a fraction of total NHEA
spending from 1.5% in the early 1960s to a fairly constant 3% since the early 1990s.
Another contributor to upstream spending is risk management. We have data from
various reports, both public (e.g., NIHCM 2002) and proprietary, that have allowed us to
assemble a partial time series on spending on drugs for treating hypertension and high
cholesterol. These data suggest that the use of these drugs grew from a negligible amount
before 1980 to at least 1.5% of NHEA spending by 2004. Risk management in total
would also include prescribed treatments for weight loss and smoking cessation, for
which we have not yet assembled the historical data. Thus, we estimate that upstream
spending has grown to more than 4.5% (=3% population-based public health + more than
1.5% risk management) of total health spending. This amount is larger than the 3%
upstream spending that was estimated in a 1991 report (Brown et al 1991). The data thus
show that upstream spending has grown as a fraction of total health spending since 1960,
even if it is still a relatively small fraction. This conclusion is significant because it
ie
stands in contrast to an impression we had before this study, that upstream spending had
in recent decades been “squeezed out” by downstream spending (Homer and Hirsch
2006).
Three balancing feedback loops have been included in Figure 5 and in our model to
indicate how, in general terms, efforts in risk management and health protection might be
funded or resourced more systematically and in proportion to indicators of capability or
relative need. Funding is not the only prerequisite for such efforts, which also depend
upon the enthusiasm and organization of the people involved (providers and patients in
the case of risk management, and the general public in the case of health protection), but
it is the leading requirement for most initiatives. Loop B4 suggests that funding for
programs promoting risk management could be made proportional to spending on
downstream care, so that when downstream care grows funding for risk management
would grow as well. Loop B5 suggests something similar for health protection,
supposing that government budgets and philanthropic investments for health protection
could be set in proportion to recent health care spending. Loop B6 takes a different
approach to the funding of health protection, linking it not to health care spending but to
risk prevalence (the stock which health protection most directly seeks to reduce). The
linkage to risk prevalence can be made fiscally through “sin taxes” on unhealthy items,
such as cigarettes (already taxed throughout the U.S. to varying extents; see Lindblom
2006) and fatty foods (Marshall 2000). In theory, the optimal magnitude of such taxes
may be rather large in some cases, as the taxes can be used both to discourage unhealthy
activities and promote healthier ones (O’ Donoghue and Rabin 2006).
Presented in Figure 6 are results from simulations in which we ask how much the
prevalence and burden of disease might have been diminished relative to the base run
(through the year 2010) if greater upstream efforts at risk management or health
protection had been made starting in 1980. These results may also be compared to a
scenario, No obesity up, in which we assume that the base run’s exogenous increase in
the onset of risk by 35% from 1980 to 2005—representing a host of socioeconomic
factors that have led to greater net caloric intake and obesity—had never occurred.
In these simulations we make assumptions about the degree to which upstream spending
can affect rates of the onset of risk, reduction of risk, and onset of disease. In particular,
we assume that maximum risk management could reduce the onset of disease by 40% and
enhance reduction of risk by 40%, and that maximum health protection efforts could
reduce onset of risk onset by 50% and enhance reduction of risk by 50%. For risk
management, the assumptions, although uncertain, have been informed by studies
focusing on the cost-effectiveness of risk management for patients with diabetes
(CDC/DCEG 2002, Hayashino et al 2004). For health protection, our assumptions are
more uncertain, because relatively little is known about the required cost and potential
impact of measures that could prevent the onset of various risk factors for disease;
somewhat more is known in this regard about preventing smoking than about preventing
obesity (Tengs et al 2001, Homer, Milstein et al 2006). The Tengs analysis, focusing on
a school-based anti-smoking program for young teens, estimated a cost of about $50 per
=i hie
student per year and projected long-term benefits in terms of reduced medical costs and
increased quality-adjusted life years.
Because our assumptions about upstream efforts are associated with significant
uncertainties, we do not purport here to provide accurate cost-effectiveness estimates,
only to illustrate how such estimates may be generated by our model. It is interesting to
ask not only to what degree upstream efforts can improve health but also whether, and
over what time frame, increases in upstream spending can be justified in terms of
subsequent reductions in downstream spending. The model calculates upstream spending
as the sum of risk management and health protection spending, and calculates
downstream spending as the sum of all personal health care spending less spending on
risk management. These measures of spending are accumulated over time, starting in
1980, as a way of quantifying overall costs and benefits; in the current model, no
discount rate is applied to these cumulative measures.
No obesity up: This simulation is presented as a “best case” alternative history to the
base run. Relative to the base run, the fraction of the population with disease grows more
slowly during the 1980s, and this fraction declines from the 1990s onward rather than
continuing to grow as in the base run. The result is much more progress starting in the
late 1980s in reducing the rate of urgent episodes, as well as a significant slowing in the
growth of health care costs. This simulation indicates the extent to which increasing risk
prevalence has undermined progress on health and has pushed health care costs upward
since the late 1980s.
More risk management: In this simulation, the strength of the assumed linkage between
personal health care spending (specifically, the nonurgent portion of that spending) and
risk management is doubled relative to the base run; thus, the strength of Loop B4 in
Figure 5 is doubled. By 2010, upstream spending per capita is increased by $108 relative
to the base run (see Figure 6), and the effectiveness of risk management (in terms of
reducing disease incidence and enhancing risk reduction) is increased to 51%, versus
27% in the base run. The increase in risk management leads to slower growth of disease
prevalence starting in the late 1980s. But, with the onset of risk left unaddressed in this
scenario, disease prevalence does grow rather than decline. Urgent episodes and health
care costs are somewhat improved relative to the base run, but not dramatically so. By
2010, a cumulative additional $359 billion in upstream spending since 1980 has led to a
reduction in downstream spending of $1,140 billion. The increased spending in risk
management is not paid back immediately, however. Not until 1995, 15 years after the
policy is initiated, does the cumulative reduction in downstream spending exceed the
cumulative increase in upstream spending.
More health protection: In this simulation, health protection is much enhanced through a
proportional funding program, starting in 1980, that devotes $5 to health protection for
every $100 of personal health care spending; thus, Loop BS in Figure 5 is activated. The
result is an immediate $90 per capita increase in upstream spending in 1980, which grows
to a $203 per capita increase by 2010 relative to the base run. By 2010, the effectivene:
of health protection (in terms of reducing risk incidence and enhancing risk reduction) is
=}
increased from its baseline value of 19% to 48% of its assumed potential. The increase in
health protection goes a long way but does not quite offset the adverse socioeconomic
influences (such as changes in food and activity environments) that would tend to
increase the onset of risk as in the base run. As shown in Figure 6, disease prevalence in
this simulation grows only a bit more beyond its 1980 level, but still does not decline as
in No obesity up. In fact, in regard to disease prevalence, urgent episodes, and personal
health care costs, this simulation produces improvements just about halfway between the
base run and No obesity up. By 2010, a cumulative additional $1,288 billion in upstream
spending since 1980 has led to a reduction in downstream spending of $2,750 billion.
The breakeven year does not occur until 2002, however, 22 years after the policy is
initiated.
The 22 year payback period under More health protection is notably greater than the 15
year payback period in the More risk management simulation. Much of this additional
payback time comes from the fact that health protection acts further upstream than risk
management does, as seen in Figure 5. Some of the additional payback time for health
protection is also likely a reflection of the fact that there is much more early (1980s)
upstream spending under the health protection scenario than there is in the risk
management scenario. The 1980s spending is arguably less cost-effective than spending
is during the 1990s, the period of most rapid growth in disease prevalence. As a partial
test of this idea, we have performed another simulation in which the health protection
program is implemented not in 1980 but in 1985. By 2010, a cumulative additional
$1,209 billion in upstream spending has led to a reduction in downstream spending of
$2,250 billion, with breakeven occurring in 2004, 19 years after the policy is initiated.
Thus, the payback period is a few years less (19 vs. 22) in this simulation, but the
breakeven year has actually been pushed back further (2004 vs. 2002).
In any event, whether the approach to upstream action is risk management or health
protection, the model suggests that the payback time, purely in terms of health care costs,
may be a relatively long one. It should be noted, however, that our model does not
include losses in productivity to employers and society at large. Another SD model
suggests that when these losses are taken into account, the payback on upstream action
may shrink to a much shorter time period (Homer, Hirsch et al 2004), a length of time
that may be acceptable to the public as well as to those employers in a position to put
upstream efforts into effect.
A couple of broad conclusions may be drawn from the model simulations presented here
in Figures 4 and 6. First, we see that cost-containment measures in the U.S. have thus far
been futile, and they have done more to limit growth in the extent of care than they have
to limit costs. In this sense, the existing market for health care services has been
dysfunctional, and it would appear that societal measures to stabilize and simplify this
market might be considered. Second, we see that progress is possible even within the
current system to reduce costs and improve health through increased investments in
upstream risk management and health protection measures. The financial payback on
such investments may take some years but could ultimately be very large.
=
Conclusion
We have sought to explain why chronic illness is such a difficult problem to deal with
and why the U.S. in particular has stumbled in both producing better health outcomes and
controlling the cost of illness.’ Part of the problem comes from the growth in health risk
that leads to greater incidence of disease, as exemplified by the rise in obesity. But
another aspect of the problem is that progress in improving the treatment of existing
illness seems to have stalled in recent years. Growth in health care ‘ts has historically
been a key driver of improved extent of care, but with rising costs, reimbursement has
become constrained, thereby creating uncertainty in the minds of potential investors
regarding future revenues, and slowing investment in assets. Another driver of extent of
care is insurance coverage, and private coverage, like reimbursement, has declined in
response to rising costs. However, the impact of such decline has been mitigated by the
availability of the government’s Medicaid program as fallback coverage for many lower-
income workers.
If rising costs are a great stumbling block to progress, why do they keep rising? Some of
the increase is simply a reflection of past progress; namely, growth in health care assets,
and increased longevity for those with chronic illness due to those health- and life-saving
improvements. If this were the whole story, one could view the recent slowing in asset
growth as part of an orderly process by which the health care system moves toward an
acceptable maximum spending level relative to GDP. However, the data show that health
care costs have continued to rise rapidly even as asset growth has slowed. The
explanation for this continued rapid growth in costs may lie in income-maintaining
adaptations by providers, who have been able to raise prices and service volumes for a
given quality of care, especially in the less well regulated outpatient sector. These
adaptations have come in direct response to the attempts by payers to control costs
through restrictions of reimbursement.
This tug-of-war between payers and providers, permitted by the current system of
payment in the U.S., has had damaging effects in terms of limiting the supply of care, and
it has also increased the administrative overhead of providers. In the absence of effective
controls, health care cost: fraction of GDP in the U.S. accelerated ahead of those in
other industrialized countries starting in the 1980's, without delivering better care
(Docteur and Oxley 2003). In this sense, the entrepreneurial U.S. health care system
which made such great progress in the past has now become bloated and inefficient. As
many have come to realize, the time is overdue for a fundamental change in this
dysfunctional system.
The difficulty of controlling costs and improving outcomes in the U.S. suggests the need
for an innovative approach to health reform, one that emphasizes upstream efforts to
reduce the health risks that may lead to chronic illness. While spending on population-
based health protection and risk management programs has grown somewhat, it still
represents a small fraction of total U.S. health care spending. Our model suggests that
policies that shift the balance toward more upstream programs can have beneficial
mie
impacts on both health care costs and the population’s health status. Although such
upstream investments may take several years or even decades to come to fruition, it is
important to recognize that improved health is a chief aspiration of all people and
therefore deserves a commitment to strategies that will benefit both current and future
generations.
Disclaimer
The findings and conclusions in this report are those of the authors and do not
necessarily represent the views of the Centers for Disease Control and Prevention.
Notes
" Despite the clear importance of cost controls in the evolution of U.S. health care since the
1970s, we have no hard data on how they have changed over time. Consequently, although
reimbursement and coverage restrictions play a central role in Figure 2, they are not included
explicitly in the simulation model. Instead, in the model we focus on personal health care costs
as a fraction of GDP as the key factor to which other variables in Figure 2 react, including the
reinvestment rate in new assets, private insurance coverage, and inflation in health care prices.
Each one of these responses to increased costs is modeled with a delay of 3 or 4 years, reflecting
the adjustment times of the relevant stakeholders, who (depending upon the variable in question)
may be employers, prospective investors, or health care providers.
* We have modeled the fraction covered by any insurance (“total”) as the sum of (1) the fraction
covered by private insurance and (2) the fraction covered by government insurance but not by
private insurance. (A large fraction of the elderly have both Medicare and private insurance to
supplement Medicare’s copay requirements and gaps in coverage. Because this population has
both public and private insurance, one cannot simply add up the different categories of insurance
to get the total covered population.) In line with the “fallback” argument in the main text, the
government-only coverage is modeled as a fraction of those not covered by private insurance, a
fraction specified by an exogenous time series. Based on Census data from 1987-2005, we
estimate that this fallback fraction has risen only slightly, from 48% in the late 1980s to 50% by
2000. For the years preceding 1987, we have examined the NHEA which provides a breakdown
of healthcare spending by major payer category, to see how the balance of private and
government-paid spending has changed over time. Based on these data in conjunction with the
Census data for the late 1980s, we estimate that the fallback fraction sat at about 18% during the
years preceding the establishment of Medicare and Medicaid in 1965 but rose rapidly thereafter to
37% in 1970, 45% in 1975, and then gradually up to 48% in 1990.
. Although Loops B2 and B3 appear to act similarly, their impacts on disease management or
urgent care, as subsets of disease care, are rather different. With regard to Loop B3, we have
noted previously that the uninsured poor have greater access to urgent care, through hospital
emergency departments, than they do to disease management. Thus, when a person loses
insurance coverage, this will tend to lead to more of a reduction in disease management than in
urgent care. Lacking disease management, this person becomes more prone to complications of
disease leading to expensive hospitalization. The net result may be not so much a reduction in
health care costs for that person as a shifting of those costs from the employer to hospitals and the
general public. With regard to Loop B2, the trend in reduced investment appears to have been
SS
more benign, more genuinely cost saving. Our model-based analysis of NHEA data suggests that
the fraction of investments directed to disease management rather than urgent care was roughly
30% through 1980 but by 2004 had increased to 35%. That is, investors appear to have moved
gradually more in the direction of disease management and away from urgent care. This could be
either because investors recognize that health care payers have become more interested in disease
management during an age of cost containment or because investors are betting that future
technological opportunities lie increasingly in disease management.
4 Simulation beyond 2005 requires assumptions for several different input time series. For future
deaths and total population, we have used U.S. Census projections. For future growth in real
GDP per capita, we have assumed a flat 1.9% per year, which was the average growth rate for
1985-2005. For future changes in risk onset and disease onset due to social and economic
influences beyond the scope of the model, we have assumed no further increase after 2005. For
the government coverage fraction of people not covered by private insurance plans, we have
assumed no further change beyond 2000, at which time we estimate the fraction at 50%. And for
risk management spending as a fraction of total disease and risk management spending, we
assume linear growth extrapolating from the past: from 1.5% in 1989 to 3.1% in 2004 to 3.8% in
2010.
> The world’s developing countries face significant growth in chronic illness prevalence
(Mackenbach 1994, Olshansky and Ault 1986, Wild et al, 2004, Mathers and Loncar 2006) and
lack the resources to duplicate the expensive patterns of care that emerged in the US. These
countries will need to find their own path. One study indicates that developing countries will
require very different prevention strategies for cardiovascular diseases than those of higher
income countries (Reddy and Yusuf 1998). The authors suggest that the relatively low levels of
the conventional cardiovascular disease risk factors in the large rural segments of the developing
countries offer a window of opportunity for early and effective control of the epidemic. They
state: “At the present levels of these risk factors in the developing countries, the approach would
be predominantly non-pharmacological, population based, and lifestyle linked. This would largely
avoid the biologic and economic costs of a pharmacological approach warranted by high levels of
these risk factors in the developed countries.”
References
American Diabetes Association (ADA). Economic costs of diabetes in the U.S. in 2002. Diabetes
Care 2003; 26(3):917-932.
Anderson GF, Hussey PS, Frogner BK, Waters HR. Health spending in the United States and the
rest of the industrialized world. Health Affairs 2005; 24(4):903-914.
Brown R, Elixhauser A, Corea J, Luce B, Sheingod S. National expenditures for health promotion
and disease prevention activities in the United States. Battelle Medical Technology Assessment
and Policy Research Center, Washington, DC. Report BHARC-013/91-019; 1991.
Bureau for Labor Statistics (BLS). Consumer Price Index — All Urban Consumers (Series ID’s
CUURO000SA0 and CUURO000SAM). _ http://data.bls.gov/cgi-bin/surveymost?cu. Last
accessed 24 January 2007.
mith «
Centers for Disease Control and Prevention (CDC 1999). Ten great public health achievements
— United States, 1900-1999. MMWR 1999;48(12):241-3.
Centers for Disease Control and Prevention (CDC 2006). Health protection goals. Accessed
October 5, 2006 at: http://www.cdc.gov/about/goals/goals.htm
CDC Diabetes Cost-Effectiveness Group (CDC/DCEG). Cost-effectiveness of intensive
glycemic control, intensified hypertension control, and serum cholesterol level reduction for Type
2 diabetes. JAMA 2002; 287(19):2542-2551.
Centers for Disease Control and Prevention (CDC/BRFSS). Behavioral Risk Factor Surveillance
System. Prevalence Data — General Health.
http://apps.nccd.cdc.gov/brfss/list.asp?cat=HS & yr=2005&qkey=1100&state=All. Last accessed
24 January 2007.
Centers for Disease Control and Prevention (CDC/NHIS). National Health Interview Survey,
Series 10 (annual). Respondent-Assessed Health Status and Reported Chronic Conditions.
http://www.cdc.gov/nchs/products/pubs/pubd/series/sr10/ser10.htm. Last accessed 24 January
2007.
Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and
Health Promotion (CDC/NCCDPHP 2007a). Chronic Disease Overview; 2007. Available at
http://www.cdc.gov/nccdphp/overview.htm
Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and
Health Promotion (CDC/NCCDPHP 2007b). Overweight and Obesity: Obesity Trends; 2007.
Available at http://www.cdc.gov/nccdphp/dnpa/obesity/trend/index.htm
Centers for Disease Control and Prevention (CDC/NVSR). National Vital Statistics Reports,
Volume 54, Number 13; 2006. http://www.cdc.gov/nchs/data/nvsr/nvsr54/nvsr54_13.pdf. Last
accessed 24 January 2007.
Center for Medicare and Medicaid Services (CMS). National Health Expenditure Data.
http://www.cms.hhs. gov/NationalHealthExpendData/02_. Last accessed 24 January 2007.
Cutler DM, Rosen AB, Vijan S. The value of medical spending the United States, 1960-2000.
New Engl J Med 2006; 355(9): 920-927.
Docteur E, Oxley H. Health-care systems: Lessons from the reform experience. Organization for
Economic Cooperation and Development (OECD), Economics Department Working Papers No.
374; 2003. Available at: http://www.oecd.org/eco.
Dominick KL, Ahern FM, Gold CH, Heller DA. Relationship of health-related quality of life to
health care utilization and mortality among older adults. Aging Clin Exper Research 2002;
14:499-508.
Eckholm E and the staff of The New York Times. Solving America’s Health-Care Crisis. Times
Books (Random House): New York; 1993.
Evans RG, Barer ML, Marmor TR. Why are Some People Healthy and Others Not? The
Determinants of Health of Populations. A. de Gruyter: New York; 1994.
mae
Eyre H, Kahn R, Robertson RM, and others. Preventing cancer, cardiovascular disease, and
diabetes: A common agenda for the American Cancer Society, the American Diabetes
Association, and the American Heart Association. Circulation 2004; 109:3244-3255.
Freudenheim M. Company clinics cut health costs. New York Times 14 January 2007.
Fries JF, Koop CE, Sokolov J, Beadle CE, Daniel W. Beyond health promotion: reducing need
and demand for medical care. Health Affairs 1998;17(2):70-84.
Gerberding JL. Protecting health—The new research imperative. JAMA 2005; 294(11):1403-
1406.
Greenlund KJ, Zheng ZJ, Keenan NL, and others. Trends in self-reported multiple cardiovascular
disease risk factors among adults in the United States, 1991-1999. Arch Intern Med 2004;
164:181-188.
Hajjar I, Kotchen JM, Kotchen TA. Hypertension: Trends in prevalence, incidence, and control.
Annu Rev Public Health 2006; 27:465-490.
Hanna KE, Coussens C. Rebuilding the Unity of Health and the Environment: A New Vision of
Environmental Health for the 21st Century. Institute of Medicine, National Academy Press:
Washington, DC; 2001.
Hayashino Y, Nagata-Kobayashi S, Morimoto T, and others. Cost-effectiveness of screening for
coronary artery disease in asymptomatic patients with Type 2 diabetes and additional atherogenic
risk factors. J Gen Intern Med 2004; 19(12):1181-1191.
Heirich M. Rethinking Health Care. Westview Press: Boulder, Colorado; 1999.
Hirsch GB, Bergan TA, Goodman MR Examining Alternatives for Improving the Nation's Oral
Health: A System Dynamics Model of the Dental Care Delivery System. A report to the Division
of Dentistry, Bureau of Health Manpower, DHEW: Washington, DC; 1975.
Hirsch GB, Myers R. Designing Strategies for Particular Health Problems: The Indiana
Cardiovascular Disease Model, Indiana Health Planning and Development Agency; 1975.
Hirsch GB, Immediato CS. Microworlds and generic structures as resources for integrating care
and improving health. System Dynamics Review 1999; 15(3): 315-330.
Hirsch G, Homer, J, McDonnell G, Milstein B. Achieving health care reform in the United States:
Toward a whole-system understanding. 23” Intl Conf SD Society, Boston, Massachusetts; 2005.
Hoffman C, Rice D, Sung H. Persons with chronic conditions: Their prevalence and costs. JAMA
1996; 276(18):1473-1479.
Homer JB. Why we iterate: Scientific modeling in theory and practice. Sys Dyn Review 1996;
12(1):1-19.
Homer JB, Hirsch GB. System dynamics modeling for public health: Background and
opportunities. Am J Public Health 2006; 95:452-458.
md
Homer J, Hirsch G, Minniti M, Pierson M. Models for collaboration: How system dynamics
helped a community organize cost-effective care for chronic illness. SD Review 2004; 20(3):199-
222.
Homer JB, Milstein B. Communities with Multiple Afflictions: A System Dynamics Approach to
the Study and Prevention of Syndemics. Proceedings of the 20th International Conference of the
System Dynamics Society, Palermo, Italy, System Dynamics Society; 2002.
http://www.systemdynamics.org/conf2002/papers/Homer|.pdf
Homer J, Milstein B. Optimal decision making in a dynamic model of community health. 37”
Hawaii Intl Conf System Sciences, IEEE: Waikoloa, Hawaii; 2004. Available at
http://csdl.computer.org/comp/proceedings/hicss/2004/2056/03/205630085a.pdf
Homer J, Milstein B, Dietz W, Buchner D, Majestic E. Obesity population dynamics: Exploring
rd
historical growth and plausible futures in the U.S. 24” Intl Conf SD Society, Nijmegen,
Netherlands; 2006.
Institute of Medicine (IOM 2001). Crossing the Quality Chasm: A New Health System for the
21" Century. National Academy Press, Washington, DC.; 2001.
Institute of Medicine (IOM 2002). The future of the public's health in the 21th century.
Washington, DC: National Academy Press; 2002.
Jones AP, Homer JB, Murphy DL, Essien JDK, Milstein B, Seville DA. Understanding diabetes
population dynamics through simulation modeling and experimentation. American Journal of
Public Health 2006;96(3):488-494.
Leonhardt D. What’s a pound of prevention really worth? New York Times 24 January 2007.
Lindblom E. State cigarette excise tax rates and rankings. Campaign for Tobacco-Free Kids:
Washington, DC; 2006. http://www.tobaccofreekids.org/research/factsheets/pdf/0097.pdf. Last
accessed 5 February 2007.
Luginbuhl WH, Forsyth BR, Hirsch GB, Goodman MR. "Prevention and Rehabilitation as a
Means of Cost Containment: The Example of Myocardial Infarction." Journal of Public Health
Policy 1981; 2(2):103-115.
Mackenbach JP. The epidemiologic transition theory. Journal of Epidemiology and Community
Health 1994;48(4):329-31.
Marshall T. Exploring a fiscal food policy: The case of diet and ischaemic heart disease. BMJ
2000; 320:301-305.
Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030.
PLoS Medicine 2006; 3(11): 2011-2030.
McKinlay JB. A case for refocusing upstream: the political economy of illness. In: Jaco EG,
editor. Patients, Physicians, and Illness: a Sourcebook in Behavioral Science and Health. 3rd ed.
New York, NY: Free Press; 1979. p. 9-25.
-24-
McKinlay JB, Marceau LD. Upstream healthy public policy: lessons from the battle of tobacco.
Int. J. Health Serv. 2000;30(1):49 - 69.
National Institute for Health Care Management Foundation (NIHCM). Prescription Drug
Expenditures in 2001: Another Year of Escalating Costs. Washington, DC., 2002.
http://www.nihcm.org/spending2001.pdf. Last accessed 24 January 2007.
National Institutes of Health, National Heart, Lung and Blood Institute (NIH/NHLBI). Morbidity
& Mortality: 2004 Chart Book On Cardiovascular, Lung, And Blood Diseases. Washington, DC,
2004. Available at http://www.nhlbi.nih.gov/resources/docs/cht-book.htm
National Safety Council (NSC). What are the odds of dying? (Odds of death due to injury, 2003).
http://www.nsc.org/Irs/statinfo/odds.htm. Last accessed 24 January 2007.
Northridge ME, Sclar ED, Biswas P. Sorting out the connections between the built environment
and health: A conceptual framework for navigating pathways and planning healthy cities. J
Urban Health (Bulletin NY Academy Med) 2003; 80(4):556-568.
O’Donoghue T, Rabin M. Optimal sin taxes. J Public Economics 2006; 90:1825-1849.
Olshansky SJ, Ault AB. The fourth stage of the epidemiologic transition: the age of delayed
degenerative diseases. Milbank Quarterly 1986;64(3):355-91.
Prentice AM. The emerging epidemic of obesity in developing countries. Intl J Epidemiology
2006; 35:93-99.
Reddy, K S and S Yusuf,, Emerging Epidemic of Cardiovascular Disease in Developing
Countries, Circulation. 1998;97:596-601
Ratanawijitrasin S. The dynamics of health care finance: A feedback view of system behavior.
Ph.D. dissertation, SUNY Albany, New York; 1993.
Simon E. Bosses push staff to eat right, exercise. Associated Press (New York) 4 December 2006.
Smedley BD, Syme SL. Promoting health: intervention strategies from social and behavioral
research. Institute of Medicine, National Academy Press: Washington, DC; 2000.
Starr P. The Social Transformation of American Medicine. Basic Books: USA; 1982.
Tengs TO, Osgood ND, Chen LL. The cost-effectiveness of intensive national school-based anti-
tobacco education: Results from the tobacco policy model. Preventive Medicine 2001; 33:558-
570.
Thorpe KE, Florence CS, Howard DH, Joski P. The rising prevalence of treated disease: Effects
on private health insurance spending. Health Affairs (Web Exclusive) 2005; W5:317-325.
U.S. Census Bureau (Census 2007a). Health Insurance Coverage Status and Type of Coverage
by Age: 1987 to 2005 (Table HI-7).
http://www.census.gov/hhes/www/hlthins/historic/hihistt7.html. Last accessed 24 January 2007.
mh =
U.S. Census Bureau (Census 2007b). Physicians by Sex and Specialty: 1980 to 2004 (Table
153), The 2007 Statistical Abstract. http://www.census.gov/compendia/statab/health_nutrition/.
Last accessed 24 January 2007.
Weisbrod BA. The health care quadrilemma: An essay on technological change, insurance,
quality of care, and cost containment. J Econ Literature 1991; 29:523-552.
Wild S, Roglic G, Green A, Sicree R, King H. Global prevalence of diabetes: Estimates for the
year 2000 and projections for 2030. Diabetes Care 2004; 27:1047-1053
Wilkinson RG, Marmot MG, editors. The Solid Facts: Social Determinants of Health. 2nd ed.
Centre for Urban Health, World Health Organization: Copenhagen; 2003.
Woolhandler S, Campbell T, Himmelstein DU. Costs of health care administration in the United
States and Canada. New Engl J Med 2003; 349:768-775.
World Health Organization. Innovative Care for Chronic Conditions: Building Blocks for Action;
2002. Available at http://www.who.int/chp/knowledge/publications/icccglobalreport.pdf
World Health Organization. Preventing Chronic Diseases: A Vital Investment; 2005. Available
at http://www.who.int/chp/chronic_disease_report/en
Yach D, Hawkes C, Gould CL, Hoffman KJ. The global burden of chronic diseases: Overcoming
impediments to prevention and control. JAMA 2004; 291(21):2616-2622
Yach D, McKee M, Lopez AD, Novotny T. Improving diet and physical activity: 12 lessons from
controlling tobacco smoking. BMJ 2005; 330:898-900.
mBhie
Risk Disease
Adverse behaviors & management management
living conditions
Population
Complications
net inflow
(urgent episodes)
Disease
Disease cure]
and recovery Disease related
deaths
Injury & violence
death rate
Urgent care
Figure 1. Population stocks and flows as modeled
-27 -
Insurance
coverage
Extent of
disease care
(Disease management, Rt Investment in
Urgent care) new assets _
= PS ft
| (a Personal a
Complications _— healthcare
spendi Reimbursement and
egal ee coverage restrictions
Be”
2 Healthcare R3
prices
i Provider
Disease é
incidence Sadana
Figure 2. Theory of downstream health care system expansion and adaptation
Key to downstream feedback loops:
RI: Health care revenues are reinvested for further growth
BI: Disease management reduces need for urgent care
R2: Disease care prolongs life and further increases need for care
B2: Reimbursement restriction limits spending growth
B3: Insurance denial limits spending growth
R3: Providers circumvent reimbursement restrictions
os
Healthcare consumption and investments per capita ($/yr) Healthcare consumption fraction of GDP
6,000, 02
1,200
ea i O15
4,000 Consumption-sim,
800
o1 Sim
2,000 Consumption-data (1) Data (3)
400,
0.05
oe ae
8 Investment-data (2) é
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Time (year) Time (year)
Healthcare price index vs general economy (1960=1) Insured fraction of population
3 1
09
, Total-sim
Sim aa Total-data (5)
1 Data (4) rivate-sim
07
Private-data (6)
0 06
196019651970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 19901995 2000 2005 2010
Tine (year) Time (year)
Fraction of population with disease or at-risk Deaths per population per year
08 0.01
Risk-data (7) sim
9
0.6 Risk-sim 0.009
Data (9)
os Disease-sim 0.008
Disease-data (8)
02 0.007
0 0.006
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Time (year) Time (year)
Figure 3. Baseline simulation and historical data
Data sources are as follows:
(1) NHEA personal health care spending, 1960-2004 annual, divided by population and
by GDP deflator (2000=1);
(2) NHEA investments in structures and equipment, 1960-2004 annual, divided by
population and by GDP deflator;
(3) NHEA personal health care spending divided by GDP, 1960-2004 annual;
eS
(4) BLS medical care CPI (1960=1) divided by general economy CPI (1960=1), 1960-
2005 annual;
(5) Census fraction of population of all ages covered by private or government health
insurance, 1987-2005 annual;
(6) Census fraction of population of all ages covered by private health insurance, 1987-
2005 annual;
(7) BRFSS fraction of adults who report having at least one of five specified
cardiovascular risk factors, 1991-1999 odd years;
(8) NHIS fraction of population of all ages who report their health as good, fair, or poor
(i.e., not excellent or very good), 1982-2004 annual;
(9) NVSR total deaths per year divided by population, 1960-1980 every 5 years, 1980-
2003 annual.
m=
Healthcare assets per capita Insured fraction of popn
8.000 1
6,000 0.95
No reimburse down
No coverage down
4,000 09
eimburse down
2,000 085
0 os
1960 1965. 1970 1975 1980 1985 1990 1995 2000 2005 2010 196019651970 1975 1980 1985 199) 1995 2000 2005 2010
Tame (year) Time (year)
Healthcare costs per capita Urgent episodes per capita
8,000 O18,
No coverage down,
6,000 0.16
4,000 od
2,000 oz
No price up
0 O.1
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
‘Time (year) Time (year)
Figure 4. Simulations exploring how reactions of payers, providers, and investors have
determined health care system behavior
mil @
Insurance
coverage
Extent of
disease care
(Disease management, Rt Investment in
Urgent care) new assets _
LOS. som “ey \
Complications = healthcare .
& Deaths Reimbursement and
Goh spending —~__» coverage restrictions
R2
= Healthcare 1)
Me Provider
Disease e
incidence Ave) adaptation
Pi Risk
i) management hes)
Be
Adverse behaviors Fi
livi iti Health protection
& living conditions funding and
organizing
Figure 5. Extending the theory to allow for upstream responses
Key to upstream feedback loops:
B4: Risk management proportional to downstream spending can help limit it
BS: Health protection proportional to downstream spending can help limit it
B6: Health protection (via sin taxes) proportional to risk prevalence can help limit it
mae
Upstream spending per capita Fraction of popn with disease
400 o4
300
nae Base
200
More risk mgmt
¥ lth protecti¢n
03
100 No obesity up
lo obesity up
0.25
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Time (year)
196019651970 1975 1980 1985 1990 1995 2000 2005 2010
Time (year)
Healthcare costs per capita Urgent episodes per capita
8,000 018
6,000 0.16
mt
4,000 nN O14
2,000 012
No obesity U
0 oO.
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Time (year)
‘Time (year)
Figure 6. Simulations exploring how upstream responses could alter health care system
behavior
mi =
