Pharmacoeconomics 2008; 26 (9): 745-751

BRIEFING PAPER 1170-7690/08/0009-0745/$48.00/0

© 2008 Adis Data Information BV. All rights reserved.

Estimating ‘Costs’ for

Cost-Effectiveness Analysis
Alec Miners
Health Services Research Unit and NICE Decision Support Unit, London School of Hygiene and
Tropical Medicine, London, UK

Abstract Since 1999, the National Institute for Health and Clinical Excellence (NICE)
Technology Appraisal Programme has been charged with producing guidance for
the NHS in England and Wales on the appropriate use of new and existing
healthcare programmes. Guidance is based on an assessment of a number of
factors, including cost effectiveness. The identification, measurement and valua-
tion of costs are important components of any cost-effectiveness analysis. How-
ever, working through these steps raises a number of important methodological
questions. For example, how should ‘future’ resource use be estimated, and is
there a need to consider all ‘future’ costs? Given that NICE produces national
guidance, should national unit cost data be used to value resources or should local
variations in negotiated prices be taken into account? This paper was initially
prepared as a briefing paper as part of the process of updating NICE’s 2004 Guide
to the Methods of Technology Appraisal for a workshop on ‘costs’. It outlines the
issues that were raised in the original briefing paper and the subsequent questions
that were discussed at the workshop.

1. Background Assembly Government. Two of these factors make

reference to the terms ‘costs’ and ‘resources’. Spe-
One of the responsibilities of the National Insti- cifically, an appraisal must take into account “the
tute for Health and Clinical Excellence (NICE) is to broad balance of benefits and costs” and “the effec-
provide the NHS in England and Wales with gui- tive use of available resources.”[1] Implicit within
dance on the appropriate use of new and existing these statements is the acknowledgement that
medicines and treatments, through its Technology healthcare resources are finite, and that judgements
Appraisals Programme. Indeed, guidance from the are required as to the cost effectiveness of referred
Technology Appraisal Programme constitutes a le- technologies, with only those technologies consid-
gal obligation to purchasers and providers of NHS ered by NICE to be relatively cost effective receiv-
care, with a default implementation time of ing positive guidance for use within the NHS.[2,3]
3 months following final publication. Guidance is Economic evaluation is specifically concerned
based on an assessment and appraisal of the relevant with the provision of information on cost effective-
evidence base for the technology or technologies at ness.[4] It represents a set of formal quantitative
hand. In reaching its recommendations, NICE must techniques to help decision makers choose between
take into account the factors listed in its directions competing means of resource consumption, such as
from the Secretary of State for Health and the Welsh the choice between different cancer treatments and
746 Miners

methods of preventing HIV. A core component of implies that all categories of cost should be included
any economic evaluation is an estimation of the (i.e. irrespective of whose responsibility it is to pay
mean expected costs of each treatment option or for the costs).
scenario under appraisal, where a cost is defined as Morris et al.[4] stated that there is unquestionably
the value of resources (such as staff time and medi- a link between the choice of perspective and the
cations) consumed by undertaking some activity. choice of preferred method of economic evaluation.
Typically this would include not only the input costs For example, cost-benefit analysis is firmly rooted
associated with each option (such as the costs of in welfare economics, and emphasises the impor-
drugs for people with HIV) but also cost savings tance of a ‘societal’ perspective in terms of the costs
through the prevention of ill health (such as AIDS- and benefits to be included in an evaluation. This
associated inpatient stays). The purpose of this pa- approach has some appeal in so much as it is theory
per is to discuss issues with respect to costing for
driven, it avoids the potential problems caused by
economic evaluation in the context of updating the
shifting costs between sectors, and it provides clear
2004 Methods Guide.[1]
guidance on which costs and benefits to include.
However, others[6] have argued that the information
2. What Are the Steps in
requirements for formally solving the whole alloca-
Estimating Costs?
tion problem across all sectors, as implied by this
Broadly speaking, there are three stages when welfarist approach, are excessively demanding.
estimating costs: (i) the identification of relevant Therefore, they propose that other forms of analysis,
resources; (ii) estimation of the resources con- such as cost-utility analysis, where benefits are re-
sumed; (iii) and valuation of these resources.[5] At stricted to health gains, and costs are restricted to the
first glance, these steps appear to be distinct and health sector, are more appropriate for economic
relatively straight forward to undertake. However, submissions to institutes such as NICE. Indeed, this
in practice, an analyst inevitably faces a number of ‘extra-welfarist’ approach was included by NICE in
issues that require some form of decision to be its 2004 description of its reference case for eco-
made, as there is typically more than one way to nomic submissions.[1] However, some provision was
proceed. For example, what is meant by relevant allowed for the inclusion of other broader categories
costs, how best to estimate resource use/analyse of cost “where they were considered to be impor-
resource use data and what is the optimal method of tant” in non-reference-case scenarios, which is in
valuation? While economic theory provides a num- many ways consistent with the two-stage approach
ber of solutions to these issues, the most ‘appropri- suggested by Brouwer et al.[7]
ate’ course of action is not always obvious. Indeed it Healthcare interventions often have far-reaching
may often depend on the decision-making body at implications in terms of their effect on future costs
which the analysis is aimed, and the rules and regu- and benefits. Indeed, Morris et al.[4] stated that the
lations under which they operate. This is almost identification of changes in resources may reveal a
certainly the case with economic submissions to very long list of resource implications and a general
NICE. On other occasions, answers are based more
‘rippling out’ effect. Johnston et al.[8] described four
on practical considerations.
categories of future cost that result from some origi-
nal (index) event, such as a treatment or a disease.
3. Identification of Relevant Resources
They stated that future costs are either related (e.g.
The perspective of the evaluation largely dictates the costs of providing a secondary regimen of anti-
the appropriate categories of cost to include. For HIV drugs for HIV-positive people and associated
example, a ‘UK NHS perspective’ would imply that hospital care) to the index disease/treatment or are
only costs to the NHS are to be included in an unrelated (e.g. a young patient who has been treated
analysis, whereas the term ‘societal perspective’ for HIV infection who develops cancer much later in

© 2008 Adis Data Information BV. All rights reserved. Pharmacoeconomics 2008; 26 (9)
‘Costs’ for Cost-Effectiveness Analysis 747

life). Johnston et al.[8] also stated that future costs guments for and against their inclusion are both
either occur in ‘years of life lived anyway’ (i.e. technical and philosophical. Restricting extensions
during the natural progression of a disease) or to life using disease-specific mortality rates is likely
during ‘additional years of life’ that result from the to underestimate increases in life expectancy and
index treatment. For example, in the absence of ignores the possibility of interventions for some
treatment with antiretroviral therapy for HIV, a per- conditions affecting other conditions. However, it is
son may live for say 2 years. However, with antire- also possible that the inclusion of future unrelated
troviral therapy they are likely to live for longer than costs may, at the extreme, lead to the abandoning of
this, say up to 15 years. In terms of assigning defini- all medical care, as ‘doing nothing’ is likely to
tions of future costs, years 0–2 would refer to ‘years appear to be the most cost-effective strategy.[12]
of life lived anyway’, while any point after this time
would relate to ‘additional years of life’. While these 4. Estimation of Resources Consumed
definitions of future costs are useful, assigning them
to specific events/circumstances can be complex. Once relevant costs have been identified, the next
For example, a tumour that develops 10 years after step is the measurement of resource use, for exam-
radiotherapy for cancer may or may not be related to ple, the number of international units of clotting
the index treatment/tumour. factor used to treat people with severe haemophilia
per year, the hours per year of medical staff time that
Once a perspective is defined, the inclusion of this involves and the proportion of applicable insti-
future related costs, both in years of life lived any- tute overheads.
way and in life-years gained, is relatively unconten- In some circumstances, resource measurement
tious in so much as they are a direct result of the can be a complex task. Even understanding and
decision to follow some course of action for an applying the relevant cost perspective can lead to
index event. Indeed, not to include these costs would difficulties. For example, it is possible that, in some
give rise to strange results for technologies such as situations people are directly liable for the costs of
prevention programmes, where treatments for dis- the health and social service care that they receive,
eases are often postponed rather than avoided.[9] whereas others with the same underlying disease or
While there is greater discussion to be had around condition may receive state funding (e.g. residential
unrelated costs that occur during years of life lived care and nursing homes). This could be for many
anyway, their inclusion or exclusion is academic, reasons, including different levels of income but
since by definition they remain identical across all also perhaps because of an under-supply of state-
treatment options. funded facilities in certain localities. Similarly, it is
The more contentious issue surrounds the inclu- also feasible that healthcare providers are liable or
sion or exclusion of the last category of future costs; only willing to cover a proportion of an individual’s
those that are unrelated and occur in years of life costs (again, for example, residential care and nurs-
gained. An example of this could be the costs of an ing homes). Measuring and including all resources
HIV patient who has responded to treatment and would potentially overestimate the costs for which
who later requires emergency treatment following a the budget holder was liable. While on the other
motor vehicle accident. The main argument for their hand, excluding resources on the basis of current
inclusion is an issue of consistency with respect to institutional arrangements might appear to be
the measure of benefit used with an evaluation.[10,11] blinkered and is potentially unethical.[4]
For example, if total mortality rates are used to Most analysts would no doubt agree that future
estimate future years of life gained, then it is consis- related resources consumed in years of life lived
tent to include the value of future related and unrela- anyway and in years of life gained should be includ-
ted resources, since both are likely to be consumed ed in an economic evaluation. However, it is less
when generating these total survival gains. The ar- clear how they should be measured. For example,

© 2008 Adis Data Information BV. All rights reserved. Pharmacoeconomics 2008; 26 (9)
748 Miners

the technology under appraisal could be a drug for up (non-censored) are representative of those who
early prostate cancer, but it is unlikely within the are not (censored). However, this does not necessa-
context of a trial that many patients will experience rily hold true when analysing resource use data. This
metastatic disease or death. It is probable that an is because different patients can accumulate differ-
economic evaluation of this drug will need to con- ent costs at different times, so two patients who are
sider the future costs of treating metastatic disease, censored at the same time, but with different aggre-
yet there are a number of treatment options for it.[13] gate costs, would ultimately be expected to have
In the context of NICE, possible solutions to this different total costs.[17] The upshot of this is that
problem include basing estimates of resource use on non-informative censoring cannot be assumed.[18,19]
existing NICE guidance (where it exists), other dis- Often there are differences between treatment
ease-specific guidelines or by collecting primary protocols in trials and standard practice, which cre-
data to establish routine clinical practice(s). Similar ates an issue around the inclusion of resource use
problems are encountered when estimating future data into economic analyses. In particular, data on
unrelated resources consumed, but conceptually this clinical effectiveness may have been derived from a
problem could be overcome if appropriate sets of randomized trial in which clinical practice in the
nationally collected statistics were published in ap- control arm differs from what could be considered
propriately age-adjusted formats. as representing ‘routine clinical practice’. For exam-
When resource use data are collected, analysis ple, the dose of a drug or the length of treatment with
typically focuses on generating mean estimates and it may differ. In such cases, the issue arises as to
associated measures of variance. However, resource whether to reflect the resources used in the trial or
use data are not synonymous with standard clinical resources that may better reflect routine clinical
data, meaning that they need to be analysed differ- practice. Including resource use based on the clin-
ently. This is because they are usually skewed, as ical trial data may under/overestimate the costs that
they are naturally bounded by a lower value of zero would be borne by the NHS. However, including
(negative resource use is not possible), while there is resource use based on standard clinical practice may
no definable upper bound. Moreover, it is not un- lead to a discrepancy between the effectiveness and
common that a small proportion of patients have cost data.
relatively large costs, meaning that the data are
further likely to be skewed. Given that it is the 5. Resource Valuation
‘mean’ cost rather than the ‘median’ that is impor-
tant to decision makers, a number of methods, such The final stage of costing involves the valuation
as non-parametric boot-strapping, have been sug- of the identified and measured resources. Economic
gested as ways of generating this information.[14,15] theory states that the true value of a good or service
is its opportunity cost, or value of the next best
Missing resource use data pose a particular con- alternative foregone as a result of using the resource.
cern since mean costs will be underestimated if they This is distinct from the idea of financial costs, the
are simply ignored within an analysis. Possible costs that are actually paid for a good or service.
methods that aim to account for missing data include Consider the value of a voluntary worker at a nurs-
restricting datasets to ‘complete cases’ or ‘filling in ing home. The financial cost of this person is zero,
the gaps’ based on data that have been collected.[16] as they are providing their services for nothing.
Resource use data are often unavailable because However, the opportunity cost is likely to be some
the duration of follow-up within a study was too positive value, given that they have forgone some
short to capture all events of interest. Such data are other activity (perhaps leisure time) in order to
usually referred to as ‘censored’. A key assumption volunteer. It follows that the appropriate values to
when analysing clinical data is that censoring is non- use for costing purposes in economic evaluations are
informative; that is, patients who are fully followed opportunity costs.

© 2008 Adis Data Information BV. All rights reserved. Pharmacoeconomics 2008; 26 (9)
‘Costs’ for Cost-Effectiveness Analysis 749

Conveniently, economic theory states that per- marginal costs. Buxton et al.[22] have argued that, in
fectly competitive markets produce prices that re- the long run, it can be assumed for resource alloca-
flect opportunity costs, as marginal costs and margi- tion purposes that all resources are variable (i.e. they
nal benefits are equal at the point of equilibrium. vary with the level of output), thus average and
Therefore, in these circumstances, valuing resources marginal costs will be equal. However, this is a
is straight forward. Less convenient is the fact that strong assumption that may obscure important is-
most markets are imperfect (e.g. where positive sues relating to technical efficiency[4] and ultimately
externalities exist) or simply do not exist, such as for cost effectiveness.
voluntary work. Where markets are known to be It is also sometimes the case that technologies are
imperfect, prices can be adjusted to better reflect purchased at a value less than the nationally/publicly
opportunity costs, although how these adjustments available unit cost (i.e. at a ‘discounted’ unit cost),
should be calculated in practice is less clear. Having leaving the analyst with a choice of different input
said this, it has been argued that where fixed prices values for the same resource. Discounts are general-
exist, such as for health-related groups (HRGs; in ly either locally (e.g. for a specific hospital) or
the UK, they are calculated by averaging individual nationally available to purchasers of health care
unit costs reported by UK trusts for groupings of within the country. For example, recombinant clot-
clinically similar procedures), they should not be ting factors for the treatment of haemophilia prod-
adjusted even if considered to inaccurately reflect ucts are purchased centrally by the Purchasing and
opportunity costs, on the grounds that they are the Supplies Agency of the NHS for use across the NHS
prices that are paid by purchasers of care.[4] Where and therefore discounts are received nationally. ‘Re-
markets do not exist, such as for in-house diagnostic fund schemes’ are related to discounted costs in so
tests, shadow pricing can be undertaken, but as much that they effectively reduce the average cost of
Morris et al.[4] stated, this often requires ‘heroic treatment; for example, schemes in which the manu-
assumptions’ meaning that the estimated price may facturer of a product refunds the drug costs asso-
not be an accurate reflection of the resources ‘true’ ciated with the first 3 months of treatment if a
value. condition has not responded to treatment after this
A related but further problem with HRGs is that time. There are arguments for and against using
by their very nature they represent an average cost of discounted costs within economic evaluations sub-
a number of clinically similar procedures. For exam- mitted to institutes such as NICE, and/or including
ple, independent costing studies have shown that the evaluations of possible refund schemes. For: they
relevant cost for a radical prostatectomy is approxi- represent the actual price paid by a healthcare pur-
mately £5600.[20] However, the mean relevant UK chaser. Against: they might not be available to all,
NHS reference cost is approximately £4000.[21] might not be permanent and might not be publicly
Thus, using a value of £4000 within an economic disclosable. Thus an institute would be unable to
evaluation would underestimate the ‘true’ unit cost communicate the effective price at which the prod-
of each radical prostatectomy and potentially lead to uct should be purchased and/or the contractual ar-
an erroneous recommendation. rangements that should surround it.
The importance of marginal costing is empha- NICE is often required to investigate the clinical
sized within economic evaluation. Marginal costs and cost effectiveness of more than one product at a
are calculated by assessing the change in total costs time, for any single clinical condition. Often these
as a result of producing one more unit of output. products are considered to be so closely related in
However, it is common for economic evaluations to terms of their pharmacology, and/or evidence sug-
reference average costs, which are equal to total gests that they are sufficiently clinically equivalent,
costs divided by the output quantity, as they are that they are considered as a class and a single mean
more likely to be readily available compared with estimate of cost effectiveness is generated. How-

© 2008 Adis Data Information BV. All rights reserved. Pharmacoeconomics 2008; 26 (9)
750 Miners

ever, as it is rare that they have identical unit costs, proportion of the costs of care for relatively less
consideration of cost effectiveness remains impor- wealthy individuals.
tant. A third topic highlighted in the briefing paper and
discussed at the workshop focused on the inclusion
of future healthcare and PSS costs, particularly re-
6. Discussion
garding how future related healthcare/PSS resource
use be measured, for example, using existing NICE-
This paper was originally written to inform the or other (disease-specific) guidance. Questions were
update of the NICE 2004 Methods Guide.[1] The also raised about the desirability and practicality of
process of updating the guide involved a number of distinguishing between future related and unrelated
steps, including stakeholder workshops at which costs.
briefing papers for specific methodological topics The final question related to the level of detail
were discussed. This paper is based on the briefing that was required for the updated Methods Guide
paper presented at the workshop for ‘costs’. The with respect to the analysis of patient-level resource
briefing paper outlined and raised a number of ques- use/cost data, given that the 2004 Methods Guide
tions for discussion at the workshop. Questions that contained little useful information on the subject.
were outlined in the briefing document included
preferred approaches to estimating unit cost and 7. Conclusions
resource use data. For example, should NICE’s ref-
erence case support the use of publicly available The identification, measurement and valuation of
national NHS tariff data such as the national refer- relevant costs are important components within any
ence costs, and if so, whether there should be any economic evaluation. NICE’s 2004 Guide to the
exceptions to this rule? In the event that this view Methods of Technology Appraisal outlined its pre-
was broadly supported, a follow-on question asked ferred approach to these issues in order to promote
which source(s) of information should be used to the consistency of methodological approach across
estimate unit costs that are not available through a submissions to the institute. This paper has de-
national system. The role of discounted unit costs in scribed a number of key discussion points that were
reference and non-reference case analyses were also had with respect to ‘costs’ in the context of updating
highlighted and discussed. the guide.
While the question of the ‘appropriate’ perspec-
tive for economic submissions to NICE’s Technol- Acknowledgements
ogy Appraisal Programme was considered to be an This paper was initially prepared as a briefing paper for
extremely important issue, it was not judged to be an NICE as part of the process of updating the Institute’s 2004
appropriate topic for debate in the context of updat- Guide to the Methods of Technology Appraisal. The work
was funded by NICE through its Decision Support Unit,
ing the methods, except for the wording surrounding
which is based at the universities of Sheffield, Leicester,
non-reference-case analyses, specifically, the cir- York, Leeds and at the London School of Hygiene and
cumstances in which non-reference-case costs Tropical Medicine.
should be included in economic submissions. Dis- The author has no conflicts of interest that are directly
cussions also focused on how costs should be pro- relevant to the content of this article.
portioned/attributed in cases where the NHS/Per-
sonal and Social Services (PSS) is only partially References
responsible for a cost, either for a specific individual 1. National Institute for Health and Clinical Excellence. Guide to
the technology appraisals process. London: NICE, 2004 [on-
or across different population groups. An example line]. Available from URL: http://www.nice.org.uk/niceMedia
of the latter could be if the costs of residential care /pdf/TAP_Methods.pdf [Accessed 2007 Jun 4]
2. Devlin N, Parkin D. Does NICE have a cost-effectiveness
were dependent on individual’s levels of wealth, threshold and what other factors influence its decisions? A
with the healthcare purchaser covering a greater binary choice analysis. Health Econ 2004; 13 (5): 437-52

© 2008 Adis Data Information BV. All rights reserved. Pharmacoeconomics 2008; 26 (9)
‘Costs’ for Cost-Effectiveness Analysis 751

3. Dakin HA, Devlin NJ, Odeyemi IAO. ‘Yes’, ‘No’ or ‘Yes, but’? 15. Briggs A, Gray A. Handling uncertainty in economic evalua-
Multinomial modelling of NICE decision-making. Health Pol tions of healthcare interventions. BMJ 1999; 319 (7210):
2006; 77 (3): 352-67 635-8
4. Morris SN, Devlin N, Parkin D. Economic analysis in health
16. Briggs A, Clark T, Wolstenholme J, et al. Missing, presumed at
care. Chichester: John Wiley and Sons, Ltd, 2007
random: cost analysis of incomplete data. Health Econ 2003;
5. Drummond MF, Sculpher MJ, Torrance GW, et al. Methods for
12 (5): 377-92
the economic evaluation of health care programmes. 3rd ed.
Oxford: Oxford University Press, 2005 17. Drummond M, McGuire A. Economic evaluation in health care:
6. Claxton K, Sculpher M, Culyer AJ. Mark versus Luke? Appro- merging theory with practice. New York: Oxford University
priate methods for the evaluation of public health interventions Press, 2001
[research paper]. York: University of York, Centre for Health 18. Raikou M, McGuire A. Estimating medical care costs under
Economics, 2007 Nov [online]. Available from URL: http://
conditions of censoring. J Health Econ 2004; 23 (3): 443-70
www.york.ac.uk/inst/che/pdf/rp31.pdf [Accessed 2008 Jul 11]
7. Brouwer WB, van Exel NJ, Baltussen RM, et al. A dollar is a 19. Young T. Estimating mean total costs in the presence of censor-
dollar is a dollar: or is it? Value Health 2006; 9 (5): 341-7 ing: a comparative assessment of methods. Pharmacoeconom-
8. Johnston K, Buxton MJ, Jones DR, et al. Assessing the costs of ics 2005; 23 (12): 1229-42
health care technologies in clinical trials. Health Technol As- 20. Calvert NW, Morgan AB, Catto JW, et al. Effectiveness and
sess 1999; 3 (6): 1-76 cost-effectiveness of prognostic markers in prostate cancer. Br
9. Gold MR, Siegel JE, Russell LB, et al. Cost-effectiveness J Cancer 2003; 88 (1): 31-5
analysis in health and medicine. New York: Oxford University
21. UK Department of Health. NHS reference costs 2006-07.
Press, 1996
London: DH, 2008 Feb [online]. Available from URL: http://
10. Mushlin AI, Fintor L. Is screening for breast cancer cost-
effective? Cancer 1992; 69 Suppl. 7: 1957-62 www.dh.gov.uk/en/Publicationsandstatistics/Publications/Pub
licationsPolicyAndGuidance/DH_082571 [Accessed 2008
11. Johannesson M, Meltzer D, O’Conor RM. Incorporating future
costs in medical cost-effectiveness analysis: implications for Mar 1]
the cost-effectiveness of the treatment of hypertension. Med 22. Buxton MJ, Acheson R, Caine R, et al. Costs and benefits of the
Decis Making 1997; 17: 382-9 heart transplant programmes at Harefield and Papworth Hospi-
12. Moskowitz M. Costs of screening for breast cancer. Radiol Clin tals [research report no.2]. London: Department of Health and
North Am 1987; 25: 1031-7 Social Security, 1985
13. National Collaborating Centre for Cancer. Prostate cancer: diag-
nosis and treatment [online]. Available from URL: http://
www.nice.org.uk/nicemedia/pdf/CG58FullGuideline.pdf [Ac- Correspondence: Dr Alec Miners, Health Services Research
cessed 2008 May 1]
Unit, London School of Hygiene and Tropical Medicine,
14. Briggs A, Gray A. The distribution of health care costs and their
statistical analysis for economic evaluation. J Health Serv Res Keppel Street, London, WC1E 7HT, UK.
Policy 1998; 3 (4): 233-345 E-mail: [email protected]

© 2008 Adis Data Information BV. All rights reserved. Pharmacoeconomics 2008; 26 (9)