Advancing the Science of Health Care Costing

Article (PDF Available)inMedical care 47(7 Suppl 1):S120-6 · July 2009with43 Reads
DOI: 10.1097/MLR.0b013e3181a9d366 · Source: PubMed
Advancing the Science of Health Care Costing
Joseph Lipscomb, PhD,* Paul G. Barnett, PhD,† Martin L. Brown, PhD,‡§
William Lawrence, MD, MS,¶ and K. Robin Yabroff, PhD‡§
he preceding articles in this volume amply illustrate and critically discuss the major
issues in health care costing. This concluding article has 2 purposes. First, we
synthesize and evaluate the main findings. Second, we identify the elements of a research
agenda for improving the scientific soundness and relevance of health cost analyses for
decision making.
As noted,
most health cost studies either assess the economic burden of disease or illness,
or contribute to the economic evaluation of specific interventions. Whatever the arena of
application, a health cost analysis generally proceeds through a sequence of steps that
should be tailored to the problem at hand.
Defining the purpose, scope of included costs, and intended audiences. Most cost
analyses are intended to inform decision making. Hence, the analyst should clearly
define at the outset the purpose, the scope (types of cost to be included), and intended
Identifying the resources used and their economic costs. The analyst must determine
how resources (labor, capital, supplies) will be attributed (or assigned) to the disease,
health problem, or interventions of interest, and how unit costs will be assigned to those
resources. The dual tasks of cost attribution
and cost assignment
constitute the
“costing out” process, and are closely related to the purpose and scope of the analysis.
Most articles in this volume focused on these essential tasks.
Deriving statistically sound conclusions about costs for the application at hand. Whether
the application is a macro-level description of disease burden,
the generation of cost
elements for a cost-effectiveness analysis,
a study of the cost difference between
interventions in a clinical trial,
or an assessment of the determinants of cost variations
across disease groups,
selection of an appropriate statistical framework and modeling
plan is essential.
Reporting cost analyses accurately, clearly, and transparently. Not only the final results,
but the important data and methods assumptions made along the way should be made
Transparency is important to both decision makers and to researchers seeking
to advance the science of health care costing.
The sections that follow examine the steps in health costing, with an emphasis on
the research agenda. Specifically, we (1) assay ongoing work to extend the boundaries of
cost analyses in the major arenas of application, (2) assess the challenges in cost
attribution, (3) discuss strategies for enhancing available data and developing new data
resources for health costing, (4) identify recent advances and some remaining issues in the
statistical analysis of cost data, and (5) consider pathways toward achieving greater
standardization in reporting and conducting health cost analyses. In each section, our
summary observations appear in italics.
From the *Department of Health Policy and Management, Rollins School of Public Health, Emory University, Atlanta, Georgia; †Department of Veterans
Affairs, Health Economics Resource Center, Palo Alto, California; ‡Health Services and Economics Branch, Applied Research Program, Rockville,
Maryland; §Division of Cancer Control and Population Sciences, National Cancer Institute, Rockville, Maryland; and ¶Center for Outcomes and Evidence,
Agency for Healthcare Research and Quality, Gaithersburg, Maryland.
The views expressed in this articles are those of the authors, and no official endorsement by the US Department of Veterans Affairs or the US Department of
Health and Human Services, including the Agency for Healthcare Research and Quality and the National Cancer Institute, is intended or should be inferred.
Reprints: Joseph Lipscomb, PhD, Department of Health Policy and Management, Rollins School of Public Health, Emory University,1518 Clifton Road NE,
Atlanta, GA 30322. E-mail:
Copyright © 2009 by Lippincott Williams & Wilkins
ISSN: 0025-7079/09/4700-0120
© 2009 Lippincott Williams & WilkinsS120 |
Important work is underway to develop better measures
of disease-specific costs at the population level and to
broaden the scope of health care costing at the individual
level, with an emphasis on costs incurred outside the health
care system.
Refining the National Accounts
Rosen and Cutler
argue that to assess whether the
nation is getting good value for money in health care, the
National Health Expenditure Accounts (NHEA) should be
augmented in ways that allocate spending to specific diseases
and health problems. Aggregate spending on each disease or
problem could then be compared with the corresponding
changes in population health status. Analytically, this in-
volves 3 major tasks: linking microlevel spending data to
national-level totals, defining the disease-based subaccount
categories, and then allocating spending to the subaccounts.
As Rosen and Cutler recognize, challenges abound in
defining subaccounts that are mutually exclusive, exhaustive,
and clinically meaningful. For example, since diabetes pre-
disposes some people to heart attacks, should the cost of a
heart attack contribute to the diabetes category, or should it be
a separate subaccount? The authors outline encounter-based,
episode-of-disease-based, and person-based approaches to this
To better understand how specific diseases and health
problems contribute to the overall burden of illness and to
facilitate evaluation of the aggregate health payoffs from
competing interventions, it is important to develop national-
level estimates of cost by disease category. Ideally, such
estimates would also be available by subpopulation and even
by major categories of disease-specific interventions.
Measuring and Valuing Productivity Costs
The NHEA focuses on expenditures incurred within the
US health care system, thus estimating aggregate direct
medical costs.
But it is frequently argued that the burden of
illness also includes lost productivity. This “human capital”
approach to measuring disease burden incorporates the value
of market and nonmarket reductions in productive activity,
arising from 2 sources. The first source is disease-related
mortality, with productivity costs based on disease-attribut-
able years of life lost and the resulting loss of productivity
over those years. The second is disease-related morbidity,
with productivity costs computed for those living with the
A recent study on the aggregate mortality-related costs
of cancer in the United States underscores the importance of
including both market and nonmarket losses,
informal caregiving and household work. In this volume,
Grosse et al
present new US data on lifetime market and
nonmarket productivity, by age and sex, which should
strengthen the empirical base for human capital-based stud-
ies. Enhancements include up-to-date, detailed data from the
American Time Use Survey on the allocation of an individ-
ual’s time across market and nonmarket activities.
However, both theoretical and ethical concerns have
been raised about the human capital approach to valuing
disease burden.
Medical expenditures and productivity
losses do not represent a complete accounting. Individuals
value good health per se. Moreover, calculations of disease
burden are substantially influenced by income levels, with
unemployed and low-earning individuals accorded a smaller
“burden” than higher-earning individuals with similar health
outcomes. Alternative approaches to valuing life and limb,
including willingness-to-pay, may be better grounded in eco-
nomic theory
and can yield dramatically different monetary
estimates of disease burden.
But to the extent that willing-
ness to pay is influenced by ability to pay, difficult ethical
issues remain. (Note, too, that some critical questions about
equity and efficiency have been raised regarding the most
common approach to deriving nonmonetary estimates of
disease burden—via quality-adjusted life years, typically in
the context of cost-effectiveness analyses.
While it has long been recognized that estimates of lost
productivity based on the human capital and alternative
approaches such as willingness-to-pay can yield substantially
different results, few comparative analyses have been per-
formed. Head-to-head studies should be conducted to assess
the magnitude of the difference in a variety of applications.
Of particular interest, in response to ethical concerns, is the
extent to which population differences in the estimated bur-
den of disease are influenced by differences in labor market
participation (a key driver of human capital calculations) and
income level (which may influence willingness-to-pay).
Putting Time into Economic Evaluations
Relatively few cost-effectiveness analyses consider the
opportunity cost of the time that individuals contribute in
securing and consuming health care services. Russell argues
that this omission biases studies in favor of interventions that
substitute patient and caregiver time for health system re-
She also points to the American Time Use Survey
as a rich source of data on time allocation to activities. A
viable alternative to survey data is to estimate patient time
costs based on service utilization patterns recorded in claims
data, medical records, and other sources.
Russell believes
that readily available average wage rates provide a reasonable
proxy for the cost of patient and caregiver time. Both com-
prehensive assessments of disease burden and estimates of
total direct cost for economic evaluations should include the
cost of time for patients and informal caregivers.
Whether the focus is on burden of illness or the eval-
uation of disease-related interventions, most analyses must
address a central issue: identifying the portion of health care
costs causally associated with the diseases or interventions of
interest over a specific time frame.
Prevalence Cost Versus Incidence Cost of
Depending on the application, the focus may be on the
prevalence cost of disease. For example, the prevalence cost
of stroke in 2008, which includes the cost of strokes occur-
Medical Care Volume 47, Number 7 Suppl 1, July 2009 Advancing Health Care Costing
© 2009 Lippincott Williams & Wilkins | S121
ring in 2008 and the continuing care of individuals in 2008
who had stroke in prior years. Or the focus may be on the
incidence cost of disease, eg, the lifetime costs associated
with all strokes occurring in 2008. Drawing from the articles
in this volume by Barlow,
Marshall and Hux,
Yabroff et al
and from the larger literature on disease
costing, we conclude that:
There is a fundamental interplay between economic and
epidemiological considerations in disease costing. Estima-
tion of prevalence cost for a given year requires, at a
minimum, data on disease prevalence for that year. This
may be obtained by direct observation of existing patients
or else derived from data on disease incidence and survival.
Estimation of incidence cost necessarily requires disease
incidence and survival. If, as likely the case, cost varies by
disease severity, information may be needed on the time
spent at each level of severity, and the costs associated with
each level. Many cost-effectiveness analyses have concen-
trated much more heavily on the epidemiological structure
of the costing model than on the resource use and cost
Studies requiring estimates of the attributable
cost of disease should focus more intensely on the cost
parameters, with no diminution of emphasis on the epide-
miological foundations for the costing model.
Aggregate prevalence cost estimates can be obtained by
multiplying directly estimated disease prevalence in a spe-
cific year by estimates of annual mean cost for those with
the disease.
But what if one wanted to predict prevalence
cost for some future year? It is possible, of course, to
simply project prevalence estimates and per person cost
forward based on trend lines. An alternative approach is to
build what amounts to a dynamic model of the incidence
cost of disease over time—which allows for secular
changes in risk factors, disease incidence, and survival—
and from this model to derive prevalence cost in any
desired future year.
In general, deriving prevalence cost from models of
incidence cost has the noteworthy advantage of promoting
coherency and internal consistency— both logically and em-
pirically— between the 2 approaches. Indeed, one can argue
that incidence costing— because it makes direct use of the
fundamental parameters of disease incidence, stage duration,
and mortality—is the “more basic” of the 2 perspectives.
The incidence cost framework could also help analysts
address a long-standing anomaly (in our view) in human
capital estimates of disease-attributable productivity loss.
While it has been common to compute morbidity costs
based on the associated productivity loss during the year of
interest, mortality costs are typically derived as the dis-
counted present-value of productivity losses in future years
arising from disease-attributable deaths in that year. A fully
dynamic model of incidence cost would make it possible
for these 2 components of productivity cost to be computed
in a time-consistent fashion, we believe.
Studies, in this volume, on the prevalence cost and inci-
dence cost of disease (focusing specifically on cancer)
amply demonstrate the insights that can be gained from
comparative analyses. Additional head-to-head cost studies
should be conducted for the major disease categories to
investigate the extent to which findings are robust to the use
of alternative data sources and costing methods.
Attributable Cost Versus Net Cost
As the articles in this volume well illustrate, many
studies require estimates of the cost that is causally associated
with a disease or disease-linked health problem of interest. In
this regard, Barlow
distinguishes between the “net” and
“attributable” cost of a disease. By net cost, he means the
observed difference between the (total) costs incurred by
individuals with the disease and a statistically similar com-
parison sample without the disease. Attributable cost refers to
those costs incurred by individuals with the disease that are
judged to be directly related to the detection or treatment of
the disease. Hence, observations on net costs are typically
obtained through an epidemiological case-control (or case-
cohort) type of study, while attributable costs ordinarily rely
on expert clinical opinion or clinical scenarios to sort out
disease-related cost from total observed cost.
Three general observations are as follows:
Obtaining an ideal representation of disease-specific costs
may be challenging under either approach. The expert
opinion or clinical scenarios used to infer attributable cost
may fail to identify all relevant disease-related procedures
or services. The causal connections between diagnosis and
subsequent costs are complex. Moreover, it is possible that
individuals under treatment with a serious disease might
forgo routine healthcare they would otherwise receive. If
there is such a “crowding-out” effect, attributable cost
could overestimate disease-specific cost. This is because
such forgone routine care (whatever its impact on health
outcomes) represents a genuine reduction in resource use in
response to the disease of interest. Hence, if the aim is to
determine the impact of the disease on all resource flows,
the net cost approach is more suitable.
An ongoing challenge in the net cost approach is achieving
an appropriate match between patients and controls. Under-
matching can result in substantial (nonrandom) heteroge-
neity between patients and controls, biasing estimates of
net cost. Over-matching (analogous to “over-fitting” in
regression modeling) may reduce the generalizability of net
cost predictions.
Both the net cost and attributable cost approaches could
benefit from clearly articulated standards to promote the
consistent application of technically sound methods across
studies. Also needed are studies that compare cost estimates
based on the 2 approaches in a variety of applications.
The challenges of deriving valid net or attributable cost
typically increase as we move ever more distant in time
from the incidence event. As time progresses, the individ-
ual with the disease of interest accumulates coexisting
illnesses and other health problems. It becomes increas-
ingly difficult to determine what portion of observed cost is
due to the disease of interest, particularly at the end of life,
regardless of the reported cause of death.
Lipscomb et al Medical Care Volume 47, Number 7 Suppl 1, July 2009
© 2009 Lippincott Williams & WilkinsS122 |
The approaches discussed by Barlow are designed to identify
the costs expressly associated with the occurrence of some
disease of interest. Neither approach is structured to analyze
the cost implications of preventing cases of the disease from
occurring—a key requirement in cost-effectiveness analyses
of disease prevention interventions. Note in particular that
under the net cost approach above, the comparison sample’s
cost per period is essentially subtracted from the cases’ cost
per period, with this net cost further adjusted by the cases’
survival probabilities. Hence, the comparison sample’s own
expected time profile of costs, which reflects both their cost
per period and their own survival probabilities, is not relevant.
By contrast, in calculating the cost savings attributable to a
prevention intervention, a key ingredient is the algebraic
difference between the time profile of costs for the cases and
the time profile of costs for the comparison sample. (In a
standard decision tree set-up, this cost difference is weighted
by the probability of preventing the disease of interest to arrive
at the expected difference in cost attributable to the
Allocating Costs to Diseases for Individuals
With Multiple Diagnoses
Among the approaches that Rosen and Cutler
for allocating aggregate health care costs to specific diseases, the
person-based method merits particular attention. This approach
lends itself naturally to multivariable regression modeling in
which cost is a function of the disease of interest and coexisting
illnesses, comorbidities, and other factors. If one believed that 2
or more diseases affect cost synergistically, interaction terms
could be included in the model to capture this joint (and possibly
hard-to-disentangle) influence.
Such regression-based cost-of-illness analyses con-
ducted at the level of the individual are highly compatible
with, and represent natural extensions of current efforts to
estimate the prevalence or incidence cost of a particular
disease or disease complex, while controlling for coexisting
illnesses and comorbidities. However, much additional work
will be required to understand the mix and magnitudes of
disease-specific costs associated with complex health prob-
lems, such as depression, and major risk factors, such as
obesity and smoking.
The focus turns now to the primary task of health care
costing: identifying the real resources consumed, and assign-
ing an economic opportunity cost to each resource.
NHEA and Population Surveys of
Service Utilization
The NHEA collects data from providers, payers, and
multiple other sources to estimate US health care expendi-
However, the NHEA does not report health expendi-
tures by individual, provider, or employer—nor by disease, as
discussed earlier. But it does provide reliable estimates, and
future-year projections, of expenditures in the aggregate and
by various population subgroups and by type of service.
The Medical Care Expenditure Survey (MEPS) obtains
population-based information on illness, health care use, and
expenditures at the individual level from US households.
MEPS is nationally representative and repeated over time, but it
systematically evaluates only certain priority conditions (such as
diabetes) and certain important clinical information, such as date
of diagnosis, is not available in all years. Consequently, disease
incidence (and incidence costs) cannot always be ascertained,
and disease-specific prevalence estimates are derivable only for
the priority conditions. Additionally, the MEPS population-
based sample may be too small to yield reliable estimates in a
given year for some conditions (eg, colorectal cancer). MEPS
also relies on individual self-reports.
MEPS is a unique national data resource, and it is
possible in principle to address some of its limitations for cost
analyses, eg, by increasing the number of priority conditions,
adding detailed clinical information, and using multiple survey
years to enlarge sample sizes within major disease categories.
This is potentially the most accurate method of assess-
ing the cost of health interventions, especially innovative
services whose resource requirements are not readily avail-
able from secondary data sources.
But microcosting studies
can be expensive, which likely accounts for why there are
comparatively few published applications. Moreover, micro-
costing methods are insufficiently standardized, so that stud-
ies with similar purposes may elect to include (and exclude)
different kinds of costs. Accurate determination of labor costs
is frequently challenging because of difficulties in measuring
how providers spend their time. New technologies, such as
personal digital assistants, may significantly improve effi-
ciency and accuracy in data collection.
Thoughtful guidelines for microcosting should clearly
identify what costs are to be included and how they are to be
analyzed. Increased transparency will promote confidence
among decision makers that the estimates are accurate and
complete for the intended application.
Activity-Based Costing
Some hospitals and integrated health systems have
implemented Activity-Based Costing (ABC) systems that are
far more accurate than cost-adjusted charges.
These sys-
tems differ from traditional hospital cost accounting systems
by automatically incorporating detailed information on work-
load and matching it to the costs of the appropriate depart-
ment. The workload detail is then used to assign costs to
individual stays and outpatient encounters. However, ABC
systems are available at relatively few hospitals and health-
care systems, and their cost estimates may be idiosyncratic to
a particular site. Estimates sometimes may be regarded as
proprietary, useful for contract negotiations but unavailable
for research. The expense of these systems may continue to
limit their adoption by small-scale providers. (That said, the
National Cancer Institute National Community Cancer Centers
Program is presently using a tailored ABC system to assess
implementation and operating costs across the Program’s 16
geographically dispersed cancer centers. Steven B. Clauser,
personal communication, March 17, 2009).
A minimum standard for ABC should be established.
Studies are needed to determine how widely ABC systems
Medical Care Volume 47, Number 7 Suppl 1, July 2009 Advancing Health Care Costing
© 2009 Lippincott Williams & Wilkins | S123
have been adopted, how they compare in terms of data
required and methods employed, and how often their findings
inform economic evaluations. In the future, a large data set of
deidentified ABC cost estimates from a representative sample
of providers could be the basis for an improved set of
standardized unit costs.
Secondary Data Analysis
Administrative data sources, particularly insurance en-
rollment and claims data, are widely used to derive dollar-
denominated measures of resource use. Because such data are
principally used for provider reimbursement, however, they
will not necessarily convey accurate information about the
economic costs of procedures and services.
For example, charge data are widely available in the US
health care system, but raw charges are poor indicators of
economic opportunity cost. A corrective long adopted by many
analysts is to cost-adjust charges using hospital cost reports
submitting routinely to Medicare.
But such cost-to-charge
ratios embody several limitations. As hospitals increase their
gross charges to subsidize uncompensated care, cost-to-charges
ratios are diminishing, which increases the possibility of incor-
rect adjustment. Under prospective reimbursement, which is
becoming more pervasive, inpatient care is paid according to the
admitting diagnosis and outpatient payments are based on pa-
tient counts. As charges become less important in reimburse-
ment, their accuracy may tend to diminish or they may no longer
be submitted to the payer. Further, cost-adjusted charges are
rarely available to characterize outpatient care, which represents
one-half of US health care costs.
Reimbursement to providers, which represents cost
from the payers’ perspective, has frequently served as a proxy
measure of economic opportunity cost in applications to both
inpatient and outpatient care. For facility-related services, the
proxy for economic cost may be based on payments accord-
ing to the diagnosis-related group system or other approaches
to calculating reimbursement. For physician services, eco-
nomic cost is frequently approximated by payments under the
resource-based relative value system, which necessarily em-
bodies a number of assumptions about the valuation and
utilization of inputs comprising such services. A continuing
challenge is accounting for all relevant payments comprising
the proxy measure, including the copayments and deductibles
charged to patients. To the extent that payments are effec-
tively set to approximate economic costs, they may be used in
analyses reflecting a societal perspective—a base-case re-
quirement in many economic evaluations.
As measures of economic costs, both cost-adjusted
charges and reimbursement to providers (payments) require
validation. Better methods are needed to estimate the cost of
ambulatory care, especially for surgeries and procedures
that can be provided only in specialized facilities. As care
continues to shift from the inpatient to the ambulatory setting,
this issue is of growing importance.
Gross Costing
A common way of estimating the cost of health care is
to multiply information on the quantity of each service by a
standard estimate of its cost.
The strength of this method
lies with its simplicity. This is why it is widely used in
clinical trials and medical decision models.
However, clinical trials often gather data on a limited
number of measures, possibly failing to document use of
important services. Utilization may also be missed in medical
decision models, which are often based on expert opinion
about services use. A related problem is whether the unit cost
is consistent with the utilization measure. An average daily
cost of hospital care will be too low if it based only on the
facility cost and does not include the cost of physician
services. Unit costs for prescription drugs may be incorrect;
although average wholesale price is widely used, most payers
receive substantial discounts from this amount.
Research is needed on whether estimates from gross
costing are consistent with more labor-intensive costing meth-
ods and to identify, in particular, the services that explain most
of the variance in cost. A list of key measures of utilization, and
standard estimates of their unit cost, may improve gross costing
accuracy. For prescription drugs, better data are needed on
such details as the appropriate discount from average wholesale
price and the dispensing fees.
Some general conclusions follow:
By using 2 or more of these costing approaches in combina-
tion, one can enrich the empirical analysis while also cross-
validating the approaches. For example, in assigning cost to a
given hospital admission, one can compare activity-based
costing, administrative data approaches, and gross costing.
Research is needed to learn if, and how, multiple costing
approaches or data sources can be combined, or used in
concert, to enhance the information base for analyses. For
example, we should further explore drawing observations
jointly from survey data, medical records, and administra-
tive files for a richer picture of health resource use.
Given the fragmented nature of health insurance coverage
in the United States, an ongoing challenge is developing
comprehensive, population-based, patient-level estimates
of disease costs. Such estimates would encompass all ages,
types of insurance and health plans, and geographic areas.
Statistical modeling is a particularly challenging area of
health costing research, though it frequently is essential for
accurately identifying the determinants of costs and predict-
ing the cost impact of interventions and policies. The articles
by Mullahy,
Basu and Manning,
and Huang
offer a
number of recommendations for advancing the field. Based
on their work, other articles in this volume, and recent
developments in the literature, the following issues especially
merit ongoing attention:
The application of multivariable modeling to cost prediction
should be enhanced. Models should be developed that provide
disease-specific costs at the macro level, accounting for the
presence of multiple competing chronic problems and comor-
bidities. In addition, more work is needed on cost prediction in
decision modeling, specifically on developing cost estimates
Lipscomb et al Medical Care Volume 47, Number 7 Suppl 1, July 2009
© 2009 Lippincott Williams & WilkinsS124 |
that are made conditional on treatment, risk factors, and other
Alternative functional forms for cost modeling, including
semi-parametric and other flexible forms, have shown promise
in improving the accuracy of cost predictions and merit
further intensive investigation. For example, generalized lin-
ear models may provide notably more robust and reliable
estimates than traditional single-part of multipart costing mod-
els, with or without log transformation of the cost.
In addi-
tion, the Cox proportional hazards model, with the hazard
parameter defined as total cost accumulated per predefined
time period, may provide a promising focal point for such
That the functional form of the cost model
can significantly impact cost-effectiveness ratios is well illus-
trated by Hoerger,
who compares regression models with
covariates entered multiplicatively compared with additively.
Approaches to assessing the validity of cost predictions
need additional exploration. As Mullahy emphasizes,
decision maker’s perspective should generally be taken into
account when planning and executing cost analyses. This
line of thinking can be formalized: the decision maker’s
utility function should be considered when selecting the
statistical criterion (eg, minimize mean square error, min-
imize mean absolute error) for assessing the accuracy of
cost predictions. In most health modeling applications,
predictive validity is not formally assessed. When it is
through, say, split-sample techniques the validation cri-
terion is not always linked to a specific objective function.
Formal inclusion of the decision maker’s perspective is
consistent with a Bayesian-oriented, statistical decision
theory approach to estimation and prediction.
Right-censoring of cost observations is a common problem
in clinical trials and many observational studies, where a
substantial proportion of subjects may “outlive” the study’s
observational interval, with their out-year costs thus unob-
servable. In response, Huang
proposes 2 approaches, each
of which relies solely on cost and survival observations
from individuals in the original clinical trial or observational
study of interest. For a survey of additional strategies for
dealing with censored cost data, see O’Hagan and Stevens.
In many chronic disease applications, not only do a high
percentage of individuals outlive the study end date, many will
incur a high percentage of their lifetime costs during the
censored period. In response, most decision modelers, includ-
ing those conducting cost-effectiveness analyses, elect to use
cost data from multiple sources to construct a projected path-
way of lifetime cost for the statistical individual portrayed in
the model.
For example, data from a clinical trial may be
used to estimate cost of the initial treatment, while insurance
claims data can be used post initial therapy until death.
If the
(unobservable) gold standard is the cost generated by a single
sample of individuals followed over their complete life course,
how close to this cost pathway will be that estimated via such
concatenation of cost observations? The validity of this data
linkage approach to deriving synthetic estimates of lifetime
cost should be examined through simulation modeling using a
variety of data sources that can provide longitudinal obser-
vations on costs.
The preceding sections convey a basic theme: signifi-
cant progress is being made, but additional work is needed in
virtually all areas of health care costing.
Continue Strengthening the Methodology
Important efforts are underway to extend the ambit and
scope of health care costing, both at the macro (national)
level and in the conduct of economic evaluations. The un-
derlying intent is to support better targeted, yet more com-
prehensive evaluation of specific health care interventions
and investments. As noted, additional work is needed on how
best to attribute costs to specific diseases, health problems,
and interventions, and on how best to assign unit cost values
to the resources consumed. The challenges in estimating the
cost of disease “episodes” deserves more attention.
There are significant opportunities to improve the em-
pirical base for health costing by enriching the set of cost-
related variables in existing data sets, linking data sets to
obtain a more complete picture of resource use, and creating
new data sets to fill current gaps. A recurring message is the
urgent need for head-to-head comparative studies to evaluate
how cost estimates vary with the choice of data set and
statistical approach to estimation and prediction.
Identify Key Investments in Research
Resources and Data Infrastructure
Existing US data resources are not fully adequate for
the key tasks of estimating the economic burden of disease or
evaluating the cost-effectiveness of interventions. To address
these challenges, the following will prove useful:
A systematic evaluation of current federal, state, and pri-
vate sector data resources should be conducted. The pur-
pose would be to assess current capacities and deficiencies;
identify strategic data linkages, comparative studies, or
other productive synergies; and determine whether new
initiatives or greater coordination of activities would be
merited. For example, such an assessment might determine
whether sources of cost data, such as the Medicare fee
schedules, could be harmonized or even synthesized with
private-sector fee schedules. Another question might be
whether existing federal data resources maintained by a
number of agencies could be better coordinated to support
head-to-head comparative studies.
Research is needed to determine whether, and how, to capi-
talize on health informatics technologies and infrastructure to
improve efficiency, accuracy, and comparability in costing.
Toward Standardized Approaches to Health
Care Costing
As studies proceed, so that the data and methods grow
ever stronger, the stage becomes set for greater standardiza-
tion in conducting and reporting cost analyses. While we do
not recommend a particular pathway forward, we can envi-
sion a consensus process that would:
Be conducted at the Federal agency level, or as a public-
private undertaking at the Institute of Medicine or another
Medical Care Volume 47, Number 7 Suppl 1, July 2009 Advancing Health Care Costing
© 2009 Lippincott Williams & Wilkins | S125
organization well-positioned to convene a diverse group of
Embrace approaches to deliberation and decision making
that have worked effectively for other consensus efforts in
health care, eg, the US Panel on Cost-Effectiveness in
Health and Medicine.
In the spirit of the US Panel, develop one or more “refer-
ence cases” to guide the conduct of health cost analyses. As
defined by the US Panel, a reference case is a “set of
standardized practices that an analyst would seek to fol-
low…” (32, p xx). While a cost study might incorporate a
number of sensitivity analyses, it would always include a
“reference” (or base) case analysis.
Define important elements of the reference case. Standard-
ized cost subcategories are needed (eg, for direct medical,
direct nonmedical, and productivity costs). A standardized
set of diseases or inter-related disease clusters should be
defined. In addition, a standardized “cost catalog” should
be designed. Such a catalog would provide the unit cost for
a range of health care services, procedures, and products. It
would also include mechanisms to adjust for geographic
and temporal variations in input prices, as well as the
perspective of the analysis (societal, patient, payer). Fi-
nally, guidelines are needed on how to aggregate unit costs
to arrive at the (reference case) estimate.
In parallel, develop standards for reporting of health cost
analyses. Such standards would emphasize completeness,
accuracy, and transparency in communication of cost find-
ings and the underlying analyses.
In these ways, advances in the state of the science in
health costing can inform the conduct of analyses. Over time,
as the quantity and quality of health cost analyses grow, the
science itself will continue to improve—and so will the
analyses that follow. With these developments, health care
decision makers will have an ever-improving empirical base
for evaluating the burden of disease and the economic con-
sequences of health care investments.
A realistic assessment of the time horizon for fully
realizing these possibilities is itself a matter for further
deliberation. We can conclude, based on the animated and
productive discussions at the December 2007 cost workshop
and the subsequent articles appearing in this volume, that the
time is ripe for an accelerated pursuit of stronger methods,
better data, and greater standardization in health costing.
1. Lipscomb J, Yabroff KR, Brown ML, et al. Health Care Costing: Data,
Methods, Current Applications. Med Care. 2009;47(suppl 7):S1-S6.
2. Barlow WE. Overview of methods to estimate the medical costs of
cancer. Med Care. 2009;47(suppl 7):S33-S36.
3. Barnett PG. An improved set of standards for finding cost for cost-
effectiveness analysis. Med Care. 2009;47(suppl 7):S82-S88.
4. Yabroff KR, Warren JL, Banthin J, et al. Comparison of approaches for
estimating prevalence costs of care for cancer patients: what is the
impact of data source? Med Care. 2009;47(suppl 7):S64-S69.
5. Yabroff KR, Warren JL, Schrag D, et al. Comparison of approaches for
estimating incidence costs of care for colorectal cancer patients. Med
Care. 2009;47(suppl 7):S56-S63.
6. Hoerger TJ. Using costs in cost-effectiveness models for chronic dis-
eases: Lessons from diabetes. Med Care. 2009;47(suppl 7):S21-S27.
7. Marshall DA, Hux M. Design and analysis issues for economic analysis
alongside clinical trials. Med Care. 2009;47(suppl 7):S14-S20.
8. Hornbrook MC, Fishman PA. Assigning resources to health care use for
health services research. Med Care. 2009;47(suppl 7):S70-S75.
9. Mullahy J. Econometric modeling of health care costs and expenditures.
Med Care. 2009;47(suppl 7):S104-S108.
10. Basu A, Manning WG. Issues for the next generation of health care cost
analyses. Med Care. 2009;47(suppl 7):S109-S114.
11. Neumann P. Costing and perspective in published cost-effectiveness
analysis. Med Care. 2009;47(suppl 7):S28-S32.
12. Rosen AB, Cutler DM. Challenges in building disease-based national
health accounts. Med Care. 2009;47(suppl 7):S7-S13.
13. Heffler S, Nuccio O, Freeland M. An overview of the NHEA with
implications for cost analysis researchers. Med Care. 2009;47(suppl
14. Bradley CJ, Yabroff KR, Dahman B, et al. Productivity costs of cancer
mortality in the United States: 2000 –2020. J Natl Cancer Inst. 2008;
15. Grosse SD, Krueger KV, Mvundura M. Economic productivity by age
and sex: 2007 estimates for the United States. Med Care. 2009;47(suppl
16. Drummond M, Sculpher M, Torrance G, et al. Methods for the Economic
Evaluation of Health Care Programmes. 3rd ed. Oxford: Oxford Uni-
versity Press; 2005.
17. Yabroff KR, Bradley CJ, Mariotto AB, et al. Estimates and projections
of value of life lost from cancer deaths in the United States. J Natl
Cancer Inst. 2008;100:1755–1762.
18. Russell LB. Completing costs: patients’ time. Med Care. 2009;47(suppl
19. Yabroff KR, Davis WW, Lamont EB, et al. Patient time costs associated
with cancer care. J Natl Cancer Inst. 2007;99:14 –23.
20. Yabroff KR, Mariotto AB, Feuer E, et al. Projections of costs associated
with colorectal cancer in the United States, 2000-2020. Health Econ.
21. Hodgson T, Meiners M. Cost of illness methodology: a guide to current
practices and procedures. Milbank Mem Fund Q Health Soc. 1982;60:
429 462.
22. Cohen JW, Cohen SB, Banthin JS. The Medical Expenditure Panel Survey:
a national information resource to support healthcare cost research and
inform policy and practice. Med Care. 2009;47(suppl 7):S44-S50.
23. Frick KD. Micro-costing quantity data collection methods. Med Care.
2009;47(suppl 7):S76-S81.
24. Barnett PG. Determination of VA health care costs. Med Care Res Rev.
2003;60(suppl 3):124S–141S.
25. Riley GF. Administrative and claims records as sources of health care
cost data. Med Care. 2009;47(suppl 7):S51-S55.
26. Luce RB, Manning WG, Siegel JE et al. Estimating costs in cost-
effectiveness analysis. In: Gold MR, Siegel JE, Russell LB, et al, eds.
Cost-Effectiveness in Health and Medicine. New York, NY: Oxford
University Press; 1996:176–213.
27. Ayanian JZ, Chrischilles EA, Wallace RB, et al. Understanding cancer
treatment and outcomes: the Cancer Care Outcomes Research and
Surveillance Consortium. J Clin Oncol. 2004;22:2992–2996.
28. Huang Y. Cost analysis with censored data. Med Care. 2009;47(suppl
29. Lipscomb J, Ancukiewicz M, Parmigiani G, et al. Predicting the cost of
illness: a comparison of alternative models applied to stroke. Med Dec
Making. 1998;18:S39 –S56.
30. O’Hagan A, Stevens JW. On estimators of medical costs with censored
data. J Health Econ. 2004;23:615–625.
31. Neymark N, Adriaenssen I, Gorlia T, et al. Estimating survival gain for
economic evaluations with survival time as principal endpoint: a cost-
effectiveness analysis of adding early hormonal therapy to radiotherapy
in patients with locally advanced prostate cancer. Health Econ. 2002;
32. Gold MR, Siegel JE, Russell LB, et al. Cost-Effectiveness in Health
and Medicine. New York, NY: Oxford University Press; 1996.
Lipscomb et al Medical Care Volume 47, Number 7 Suppl 1, July 2009
© 2009 Lippincott Williams & WilkinsS126 |
    • "Therefore, we provide an overview of the types of costs and cost information that would optimally be collected in research on bereavement in health care settings, and then summarise relevant research findings. Analysing costs in health care is difficult for many reasons, including challenges in measuring costs, challenges in acquiring data, and challenges in defining what actual costs consist of (Lipscomb, Barnett, Brown, Lawrence, & Yabroff, 2009). Four types of costs have been identified as important in analysing the costs of preventive interventions such as bereavement programs (Foster, Porter, Ayers, Kaplan, & Sandler, 2007 ). "
    [Show abstract] [Hide abstract] ABSTRACT: Research to date on grief and bereavement in health care providers has focused on those experiences from the perspective of the individual. We propose, however, that the emotional costs of bereavement in the health care setting are also health care systems issues. This paper focuses on the emotional costs of grief and bereavement in health care providers, and on the economic costs of bereavement and bereavement care in health care settings. Evidence regarding the costs and cost-effectiveness of bereavement interventions is limited. We summarise existing relevant research and offer an overview of the types of costs and cost information that would optimally be collected in research on bereavement in health care settings. We also propose an analytic framework that could be used to systematically consider the larger picture of bereavement in health care settings, how available evidence fits into this picture, and what evidence is needed to improve care. This approach is derived from health services research. It is hoped that the proposed framework will prove useful in stimulating new research questions, and in guiding research that not only advances our understanding of the emotional and economic costs of bereavement but also improves bereavement care.
    Full-text · Article · Dec 2010
  • Full-text · Article · · Medical care
  • Full-text · Article · Jul 2009
Show more