Добавил:
Sekretar
kiopkiopkiop18@yandex.ru
t.me/Prokururor I Вовсе не секретарь, но почту проверяю
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз:
Предмет:
Файл:Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_5389_Библиотеки_им_академика_М_И_Перельмана
.pdf
110
8071 admissions
randomly selected from
20 hospitals, 50%
recording an inpatient
death, excluding:
children, psychiatry,
R. A. Elliott
obstetrics
Patient record review by
physicians and nurses to
identify: adverse events,
MRAEs
Assessment of causality
and preventability using
6-point Likert scales, and
review process with other
assessors
218 MRAEs in 204
admissions, of which 55
MRAEs and 53 admissions
were preventable
Fig. 5.1 Identication and categorisation of ADRs by Damen etal. [19]. MRAEs medication-related adverse events
Could theHarm Have Been Avoided?
(Preventability or Avoidability)
Not all ADRs are avoidable. Many ADRs are
caused by drugs such as aspirin, warfarin, or
diuretics that have potentially signicant longterm benets to patients and are recommended in
guidelines due to an evidence base for effect and
involve nely balanced decisions about the benets versus the risks in some categories of
patients (e.g. those with multi-morbidity).
Therefore, it can often be difcult when retro-
decision was wrong, when it had been based on
careful balancing of benets versus harms taking
into account evidence-based guidelines and
patient preferences. In previous work, we have
made the conservative assumption that only denitely avoidable ADRs approximate the harm
from medication errors [24], which produces
more conservative estimates of avoidable harm,
than if both possibly and denitely avoidable
ADRs are included, as in the work by the National
Patient Safety Agency (NPSA) [25, 26].
spectively reviewing prescribing decisions to
make a judgement as to the avoidability of the
harm. Many ADR studies tend to judge large
numbers of ADRs as possibly avoidable [21–23],
which are likely to include many cases where
hindsight bias might suggest the prescribing
Extrapolating Estimates ofHarm
fromMedication Errors
Due to the complexities of linking medication
errors to harm, another method that has been
used to generate estimates of harm from medica-

5 Economics ofMedication Safety, withaFocus onPreventable Harm
111
tion error data is to extrapolate or estimate the
harm from medication error data.
Many studies have used the concept of ranking errors by some subjective judgement of harm
severity [18, 27–30]. Apart from the NCC MERP
system reported earlier in this chapter, many
other systems have been developed, a few of
which are cited here [31–33]. One system developed by Dean and Barber [31] divides errors into
‘minor’, ‘moderate’, or ‘severe’. Thirty healthcare professionals from 4 UK hospitals scored 50
medication errors in terms of potential patient
outcomes on a scale of 0–10, where 0 represented
a case with no potential effect and 10 a case that
would result in death. Similarly to the approach
used by Damen et al. [19], limitations of this
approach lie with the intrinsic subjectivity of the
method and the fact that many studies develop
their own severity assessment system, limiting
the comparability of results from different
studies.
Another way to generate estimates of harm
from medication error data is to build a model
reecting the types of harm that might occur, and
then attach a probability of that harm, and the
impact of that harm on health and cost. A model
is a simplied version of reality that represents
input, process, and outcomes of an event and is
used extensively in health economics.
One example of where this has been carried
out is the case of NSAIDs. The side effects or
adverse events, particularly GI bleeding, renal
dysfunction, and impact on cardiovascular function, are well-recognised in NSAIDs [34–36].
We know that an NSAID can cause a range of
GI harms when given, ranging from mild intermittent GI discomfort, persistent serious GI discomfort, gastric ulcer to serious or even
life-threatening, upper and lower GI bleeding.
These events can even occur in people when an
NSAID is used appropriately, but are more
likely in people with risk factors such as age, a
previous GI bleed, long-term use or the use of
other medications that increase bleeding risk.
Prescribing NSAIDs at all, or in the absence of
gastroprotective agents (GPA), in these groups
of patients could be seen as a prescribing error.
An example of building a Markov model to estimate harm from NSAIDs is presented in Case
Study 2.
A key limitation of this approach is that not all
medications are associated with the large body of
evidence there is around NSAID harm. For
example, it might be more difcult to populate
such a model with data if one wanted to quantify
the harm associated with a drug like amiodarone
or lithium.
Measuring Harm andPatient Outcomes
fromMedication Errors
Medication errors can result in harm. Ideally, a
measure of harm would capture all the impacts
on health and patient experience, including all
effects on morbidity and mortality, over the short
and long term. This type of evidence is challenging to collect. It is not routinely available and
therefore requires specic research resource. Due
to the complexity and cost of following up a
cohort of people who have experienced a medication error, other proxies for health or reductions
in health are often used to approximate to patient
outcomes from a medication error.
The clinical indicator most commonly used to
approximate to patient outcomes after primary/
ambulatory care errors and secondary care/inpatient errors is increased use of hospital care [16].
Increased use of hospital care due to medication
errors can be measured through increased use of
emergency departments, hospital admissions,
Intensive Care Unit (ICU) admissions, length of
hospital stay or readmission to hospital or
increased level of intervention during a hospital
stay. Hospital admissions and readmissions are
the most commonly used clinical indicator to
quantify harm from errors in ambulatory/primary
care. Hospital length of stay can be used to quantify short-term harm for both inpatient medication errors and also errors in ambulatory/primary
care. Admissions to ICU, or other higher-level
care, can be used to quantify short-term harm for
both inpatient medication errors and also errors
in ambulatory/primary care.

112
R. A. Elliott
Use of hospital care as an outcome measure is
problematic because so many other factors affect
the supply and use of hospital care, including the
specic characteristics embedded in the health
system in which the studies are conducted. For
example, hospital admissions can be affected by
severity of illness, source of referral, case mix,
and socioeconomic status and as such all of these
confounders require adjustment [37].
Using Mortality toMeasure Harm
fromMedication Errors
Mortality is a useful outcome measure because it
is objective and easy to measure. Medical errors
in general are suggested to be the third leading
cause of death in the USA [38]. However, there
are problems associated with mortality as an outcome measure. First of all, people may die from
causes other than that of interest to the study,
which can mask mortality linked to the medication and thus confound the results. Second, most
medication errors reduce quality of life (QoL)
rather than mortality. Mortality is a relatively
insensitive measure that requires a study with
many patients followed up over a long period of
time. Finally, people of different ages and sex
have a different risk of mortality, so it is important that mortality comparisons correct for age
and sex. Everybody dies eventually so it is actually premature mortality we are trying to reduce,
such that individuals are able to live out an
acceptable lifespan. Therefore, if a premature
death is caused by a GI bleeding from an NSAID,
or a woman with a family history of stroke prescribed an oral contraceptive dies from a stroke,
we have lost years of life for an individual.
Therefore, mortality can be converted into lifeyears lost, or years of life lost (YLL).
Studies of medication error-related harm have
attributed deaths to medication errors. For example, in the study conducted by Pirmohamed etal.
(2004) [21], the drugs most commonly implicated
in causing hospital admissions were low-dose
aspirin, diuretics, warfarin, and NSAIDs [21]. GI
bleeding was the most common adverse effect,
occurring in 157 (72%) of all aspirin-related
admissions. Of the 28 deaths which were identied as being a direct result of the ADR (as detailed
in either the case notes or on the death certicate),
15 deaths were due to bleeds caused by NSAIDs/
aspirin/warfarin. This gave an index hospitalisation death rate of 0.15% due to ADRs (2.3% of
ADRs were fatal, 1.25% of ADRs led to fatal GI
bleeds). Osanlou etal. also reported that 0.42% of
people admitted to hospital due to an ADR died as
a result [17].
It is likely that those medication errors leading
to the most severe ADRs can lead to death.
However, it is important to be able to identify risk
of death attributable to the error. In the example
of the woman with a family history of a stroke
prescribed an oral contraceptive dying from a
stroke, she already had an underlying risk of
death from a stroke, dependent on age and other
risk factors (such as hypertension, cigarette
smoking, and migraine). The prescribing of an
oral contraceptive increases that risk, rather than
introducing a new risk [39]. It should be noted
that the quality of evidence in studies around risk
can be highly variable. A recent umbrella review
(13 meta-analyses of RCTs and 45 meta-analyses
of cohort studies) of the association of hormonal
contraceptive use with adverse health outcomes
concluded that ‘Overall, the associations between
hormonal contraceptive use and cardiovascular
risk, cancer risk, and other major adverse health
outcomes were not supported by high-quality
evidence’ [40].
Using Quality ofLife andUtility Measures
toEstimate Harm
Using clinical indicators carries the implication
that changes in these will extrapolate to an effect
on the patient’s QoL. However, there is an
increasing awareness that it is necessary to measure the impact of health care on an individual’s
QoL, or health-related quality of life (HRQoL),
at least. There are many functional, social, psychological, cognitive, and subjective factors that
impact on QoL so measuring QoL is methodologically complex. QoL measures can be disease-specic, such as the Seattle Angina
Questionnaire (SAQ) [41], or generic, such as the
World Health Organization Quality of Life
(WHOQOL) [42]. Disease-specic QoL measures tend to be more sensitive to changes in QoL

5 Economics ofMedication Safety, withaFocus onPreventable Harm
113
as they are tailored to disease-specic symptoms.
Generic QoL measures can be used to compare
QoL in people with different illnesses.
In health economics, utility is the preferred
measure, dened as the value attached by an individual to a specic level of health status or a specic health outcome. Different individuals may
attach different values to the same health state.
For example, some people may be prepared to
tolerate a lot of nausea to allow them to be pain
free. Other people may prefer to tolerate more
pain and reduce the level of nausea. The important concept here is that utility measurement
allows patients to value the health state, based on
their own preferences.
Like generic QoL measures, utility can be
used when looking at groups of patients who may
have different illnesses and can be used to compare outcomes in different patient groups. Simply,
utility is used to attach a numerical value to the
value a person has for a health state, the methods
allow individuals to indicate the direction and
strength of their preference for a particular health
state, where 1.0 indicates perfect health and zero
is equivalent to being dead. People can have utility scores below zero, if they value their health
state as being worse than being dead. Patients
complete a questionnaire describing their health
states, upon which a preference-based algorithm
is applied to estimate utilities. The EuroQoL ve
dimensions measure (EQ-5D) is the multi-attribute utility instrument most widely used to measure utilities in clinical trials and related studies
and is used most often for cost-utility evaluations
in health technology assessments [43]. These
utility measures are used to generate qualityadjusted life-years (QALYs). A QALY combines
survival periods (quantity of life) with health status valuations (quality of life) to provide a standard unit for measuring health gain. One year in
perfect health is one QALY. Less than perfect
health states such as surviving with a stroke generate fewer QALYs. Therefore, the harm associated with a medication error can be represented
as a QALY decrement, also referred to as
disutility.
The disability-adjusted life year (DALY) is a
measure of overall disease burden [44]. One
DALY represents the loss of the equivalent of 1
year of full health. DALYs for a disease or health
condition are the sum of the years of life lost due
to premature mortality (YLLs) and the years
lived with a disability (YLDs) due to prevalent
cases of the disease or health condition in a population. YLL contributes to the mortality aspect of
DALY and is the difference between standard life
expectancy and age at death. YLD accounts for
the morbidity part of DALY and indicates the
number of years of life lived in less than full
health [45]. In 2017, the total global DALY due
to adverse effects from medical care in general
was 62.79 (52.09–75.45) per 100,000 population
[46].
There are many criticisms around the design
and application of QALYs and DALYs, ranging
from the appropriateness of combining mortality
and morbidity in one measure, equity for different groups of the population (such as older adults,
people with disabilities) to tool construction and
use [47, 48], which are beyond the scope of this
chapter.
Expressing Outcomes Using Monetary
Values
Another method of measuring outcome is to convert these benets into a monetary value. The
contingent valuation or ‘willingness to pay’
method elicits monetary values for health.
Contingent valuation accounts for both health
and non-health effects and is considered by some
to be a more comprehensive measure of the
effects of health care. For example, non-health
benets may include patients’ preferences for
consideration of their dignity, or aspects associated with the process of an intervention, such as
location [49]. Willingness to pay has been used to
elicit preferences for the avoidance of side effects
with antidepressants, identifying those most troublesome to patients [50]. Blurred vision and
tremor were the side effects considered most
troublesome and were associated with the highest
willingness to pay values to avoid them. The primary concern around willingness to pay as a
method is its hypothetical nature and the valuation solicited. There are concerns that the values
are constructed in response to the questions and

114
R. A. Elliott
that they do not exist before they are measured.
This method has not often been used in the area
of medication safety.
2.1.3 Resource Use andCosts
ofMedication Errors
andAdverse Drug Reactions
A range of resources are consumed as the result
of ADRs and the interventions used to improve
safety. Health economics is concerned with identifying and quantifying all the costs incurred, i.e.
capturing the true economic cost incurred by a
medication error. Calculation of true economic
cost is difcult, but it is essential to make sure
that cost information reects true economic cost
as closely as possible. This is not usually straightforward in health care because normal markets
and pricing mechanisms are not necessarily present. This section introduces key cost concepts
and denitions and describes their use in medicines safety. The reader is also directed to
Drummond etal. (2015) [51] for further information around these methods.
Costs can be divided into direct and indirect
costs.
Direct Costs
Direct costs are the costs associated directly with
an illness or healthcare intervention. Direct medical costs are the costs incurred by the healthcare
provider and are split into xed and variable
costs. These costs include staff time, medical
supplies, hotel costs, capital costs, and overhead
costs. Also, there may be direct costs not incurred
by the health service (direct non-medical costs).
These can include:
• Patients’ out-of-pocket expenses (such as trav-
elling or child care costs)
• Costs falling on other parts of the public sec-
tor, such as social services (this would include
costs like provision of domestic help or disability pension payments)
Fixed costs are those incurred whether patients
are treated or not. The two major components of
xed costs are overhead and capital costs. Capital
costs are incurred when major capital assets such
as counselling rooms are built or equipment is
purchased. Overheads are those incurred by the
running of the service, such as lighting, heating,
and cleaning costs.
Variable costs are incurred from a patient’s
treatment. This includes disposable equipment,
drugs, blood products, investigations, and so on.
Drugs and other consumables may have prices
that vary between purchasers due to the inuence
of buying groups, contractual agreements, quantity discounts, and competitive bidding.
If a person has to stay in hospital for longer
because of an ADR, the costs associated with that
increased length of stay will be a combination of
xed costs incurred by the hospital to make inpatient beds available, allocated top-down to calculate a mean cost for 1 day’s stay, and variable
costs associated with specic interventions
needed to manage the ADR. If a person is admitted to ICU, xed costs of providing an ICU bed
will be much higher than that of a standard inpatient bed, so the xed costs are correspondingly
higher. The interventions carried out on top of
that are also likely to be much higher than on a
standard ward, so variable costs are also likely to
be higher.
How Are Costs Valued?
The two ways of collecting costs are either ‘topdown’ or ‘bottom-up’ (also called ‘microcosting’). Top-down studies use the total budget to
produce average costs per patient. This method is
the quicker one, but assumes that all patients
have the same diagnosis, severity of illness, and
treatment. The costs produced from this method
are not sensitive to changes in treatment.
Bottom-up studies measure resource use by individual patients so they are able to detect treatment differences between those patients. This
method produces much better-quality costs, but
can be time consuming and expensive to collect.
Also, top-down costs are often available from
healthcare accounting systems, whereas bottom up costs may need to be collected especially for
that study.
Resource use is measured in physical units,
such as hours of staff time or quantities of drugs
given. The next step is to attach costs to that

5 Economics ofMedication Safety, withaFocus onPreventable Harm
115
resource use. Most resource use has an associated
unit cost, such as cost per hour for staff or cost
per dose for drugs. Overheads such as heating
and lighting will also have a unit cost, sometimes
per hour, or maybe per unit area per year. Market
prices for resources are used to approximate to
the opportunity cost of that resource.
Costs Versus Charges
It is important to understand the distinction
between real cost of a healthcare intervention and
the charges that may be used to generate a price
for that intervention. The real cost is a reection
of the resources consumed during the intervention, and thus approximates to their opportunity
cost. Charges used by healthcare providers often
do not reect true cost. This is because healthcare
providers may cross-subsidise losses on some services with prots from others, at the same time
running an overall surplus on services to nance
new growth. These charges do not reect the true
economic cost of the service, although they have
an obvious operational function.
In some settings, including England [52, 53],
there are publicly available prices for medicines
and some health and social care services. These
are the prices charged for specic units of care,
so give an approximation of cost. However, many
countries, such as the USA, France, Greece, and
Cyprus, do not have a lot of readily publicly
available cost information.
Litigation Costs
From April 2015 to March 2020, National
Health Service (NHS) Resolution received 1420
claims relating to medication errors, of which
487 claims were settled with damages paid,
costing the NHS £35 million (excluding legal
costs) [54]. Another study estimated total litigation cost to the NHS was £3.6 billion in
2018/2019, a fourfold increase from 2008/9
[55]. Anticoagulants, opioids, antimicrobials,
antidepressants, and anticonvulsants were the
most common medications to be implicated in
incidents. Primary care errors that resulted in
litigation quadrupled between 2015 and 2020
with the number of claims and cost per claim
increasing every year.
An Irish study of medication-related claims
nalised from 2011 to 2016 identied that key
medication groups involved were general anaesthetics, opioids, penicillins, antithrombotics, and
local anaesthetics [56]. The most commonly
pleaded primary injuries in this study were allergic reaction, deterioration in clinical status, and
post-traumatic stress disorder. The median total
cost of these claims was €60,991, including
median damages of €33,858.
Litigation costs may be incurred by healthcare
providers as a result of ADRs, so can be considered to come under direct costs, although they are
very rarely included in studies of economic
impact of medication errors [16].
Indirect Costs
Indirect costs are incurred by the reduced productivity of a patient, and their family, resulting
from illness, death, or treatment. These may
include time out of work or other unpaid activities (such as caregiving, education); time spent
going to healthcare providers; time spent caring
for the patient by relatives or paid carers; time
forgone from leisure and other non-market activities [57]. The signicance of indirect costs
depends upon the particular illness and treatment
involved. Some indirect costs can be calculated
from data (time off work due to sick leave; early
retirement; reduced productivity at work) but
others are difcult to measure. Also, there are
unresolved issues about including indirect costs
because this would tend to favour interventions
where the individuals are in employment (i.e. not
children, ‘housewives’, the unemployed, and
older adults). The alternative to this is to attach a
value to unpaid activities, such as attending
school or carrying out housework, or informal
caregiving. Indirect costs play an important role
if the intervention produces benets that enable
the target patient group to return to work or their
normal daily activity.
Due to difculties around identifying, measuring, and valuing indirect costs, they are not often
included in economic studies both in general and
in patient safety [16]. However, it is likely that
serious ADRs will affect indirect costs, so they
should always be considered, if not measured.

116
R. A. Elliott
When indirect costs are included, they can have a
dramatic effect on the results of the analysis.
Perspective
When looking at economic burden in general, the
costs included depend on the perspective of the
analysis. Figure 5.2 summarises the different
costs included in different perspectives.
The societal perspective is the widest perspective because it looks at the costs from the viewpoint of society as a whole, including direct and
indirect costs. Different studies reporting economic burden often include or exclude some categories of costs, depending on their perspective,
and this can contribute to variations in estimates
of burden. Due to the methodological challenges
associated with collecting and valuing indirect
costs described above, many economic studies
concentrate on costs from the perspective of the
healthcare provider only, most only looking at
medication and hospitalisation costs [16]. A goodquality costing study should clearly state the perspective, along with the study population, all
relevant costs and their sources included, adjustment for differential timing of costs using dis-
counting (see section ‘Discounting Costs and
Benets’), a sensitivity analysis to address parameter or methodological uncertainties [58].
The choice of economic perspective and time
horizon are major determinants of the resources
and costs measured. A study with an acute care
hospital economic perspective and a short time
horizon will focus on the direct costs of providing hospital care for the current visit, but will not
consider costs of care after hospital discharge, or
societal costs of illness resulting from lost productivity. A signicant proportion of the cost of
ADRs is accrued after discharge from acute care,
so this study design will generally underestimate
true economic costs [59].
Discounting Costs andBenets
When determining true economic cost, adjustment of costs for differential timing (discounting) is recommended. Discounting makes current
costs worth more than those occurring in the
future because individuals would rather spend
money later than in the present [60]. This is
because any capital held now can be invested,
earning interest. Therefore, there is a cost to
Costs to social
services
Costs to primary care
Costs to secondary
care
Hospital: operating theatre,
ward, surgeon,
anaesthetist, nurse,
pharmacist, physiotherapist,
drugs, prosthesis, X-rays
etc
= hospital perspective
= health service perspective
= public sector perspective
= patient/carer perspective
= indirect costs
Fig. 5.2 Costs included across different perspectives
Primary care
visits,
medicines
= societal perspective
Domestic help,
disability
allowance,
social care
Patient out-ofpocket
expenses,
travel,
childcare
Lost
productivity

5 Economics ofMedication Safety, withaFocus onPreventable Harm
117
spending money in the present. This means that
costs incurred now are of greater importance than
costs to be incurred in the future. This is known
as a ‘positive time preference’ [61].
Discounting makes current benets worth
more than those occurring in the future because
there is desire to enjoy benets now rather than in
the future, also due to the positive time preference. Whether future health benets should be
discounted at the same rate as costs, or at all, and
whether discount rates should differ for low middle income countries (LMICs) is under constant
debate and the interested reader is directed to
Severens etal. [62] and Haackens etal. [63].
2.1.4 Evidence forEconomic Burden
ofMedication Errors
There are many estimates of economic burden of
errors, using a mixture of the methods described
above. The WHO estimates an annual cost of
€4.5–21 billion in Europe [64]. Lim et al. estimate that least 250,000 hospital admissions
annually in Australia are medication-related,
with an estimated cost of 1.4 billion Australian
dollars [65].
However, variations in methodology lead to
wide variations in estimates of economic burden
of medication errors. A fairly recent systematic
review exploring the economic impact of medication errors included 16 studies and reported that
the mean cost per error per study ranged from
€2.58 to €111,727.08, with most of this variation
being due to variation in methods used [16].
In 2007, the NPSA estimated NHS costs of
preventable medication errors to be £774 million
each year in England, at 2005/6 prices (£954 million at 2015/16 prices) [25, 26]. This sum was
derived from costs of admissions (£359 million),
costs of increased lengths of admissions from
errors occurring whilst in hospital (£411 million), and litigation costs (£4 million). A 2018
study estimated that costs to the NHS in England
of denitely avoidable ADRs leading to or
extending a hospital admission are £98.5 million
(£98,462,582) per annum, consuming 181,626
bed-days and causing or contributing to 704–
1708 deaths during the index hospitalisation [24].
The difference between the two estimates of
£954m and £98m is mostly due to the fact that
the NPSA study included harm denitely or
probably caused by medication errors and the
Elliott etal. study included harm only denitely
caused by medication errors. Neither approach is
wrong, but illustrates that the numbers in burden
of medication safety studies should be viewed
with caution.
In conclusion, the results of studies of the
economic burden of medication errors are based
on a range of methods and are probably not
directly comparable. However, they appear to
suggest that there is sufcient economic impact
to support efforts to reduce medication error
rates, associated ADRs, patient harm, and cost.
As we move on to look at the economic impact
of interventions to improve medication safety
in the next section, it is important to remember
that not all harm is avoidable. A recent metaanalysis of the types of studies described here
suggests that the pooled global prevalence for
avoidable medication harm is 3% and for overall medication harm is 9% [8]. This is essential
information because any safety intervention
can only remove avoidable harm, so in this context, can reduce harm from 9% to 6%, but never
to 0%.
3 Economic Evaluation
ofInterventions toImprove
Medication Safety
Interventions to reduce medication errors are
not new. In 1972, an educational intervention
in digoxin prescribing reduced ‘digitalis intoxication’ [66]. There have been many reviews of
these studies [12, 13, 67, 68]. Until recently,
most studies about reducing medication errors
had been undertaken in secondary care and
tend to be focused on computerised tools, educational strategies, or professional roles [12].
Strategies and initiatives that aim to change
prescribing behaviour are generally costly,
with little evidence presented around their
cost-effectiveness [11, 12]. Economic evaluation allows the decision-maker to see the
impact of their decision to fund an interven-

118
R. A. Elliott
tion, or not, on both costs and patient outcomes
and is a useful tool to inform resource allocation decisions.
3.1 What Is anEconomic
Evaluation?
An economic evaluation is concerned with identifying the differences in costs and outcomes
between options. It can be dened as a study that
compares the costs and benets of two or more
alternative interventions.
There are two main approaches to economic
evaluation: cost-effectiveness analysis (CEA)
and cost–benet analysis (CBA). They differ in
the type of outcome measure used. CEA values
outcomes in non-monetary terms so the results
are presented as incremental changes in cost per
unit of outcome. CBA values outcomes in monetary terms so the results are presented as net monetary benet (NMB). Cost–utility analysis (CUA)
is a type of CEA where utility is the outcome.
Cost minimisation analysis (CMA) is a special
case in CBA or CEA and occurs when the outcomes of different options are equivalent, so the
less costly option should be selected.
In CEA (and CUA), the following questions
are always asked:
• What is the difference in cost between the
interventions?
• What is the difference in outcome between the
interventions?
The answers to these questions allow the derivation of the incremental cost-effectiveness ratio
(ICER). ICERs may be calculated using the following equation: ICER = (C1 − C0)/(E1 − E0),
where C1=cost in intervention group; C0=cost
in control group; E1 = effect in intervention
group; E0=effect in control group.
The ICER expresses the cost required to
achieve each extra unit of outcome. When one
alternative is more effective, but requires more
resources, the ICER must be calculated. In the
situation when one alternative is more effective
and less costly, this alternative is the dominant
therapy. When there is dominance, ICERs do not
need to be generated. The generation of the ICER
allows us to see how much extra cost is incurred
for the extra benet. It is then left to the decisionmaker to make a value judgement as to whether
they think that the extra benet is worth the extra
cost. For example, in England, National Institute
for Health and Care Excellence (NICE), the
decision- making body that recommends technologies for NHS funding, has a cost-effectiveness
threshold at £20,000 per QALY.An intervention
with an ICER below £20,000 per QALY would
be considered a cost-effective use of NHS
resources. In other countries, the threshold may
not be so explicit and may vary by sector or disease area.
3.2 The Intervention
A key aspect of study design is the denition,
description, and clarity of intent of the intervention being carried out, as this will affect both
effectiveness and cost aspects of the evaluation,
as well as affecting the relevance of the evaluation to other settings. Whether the intervention is
in itself grounded in evidence needs to be presented, and as many safety interventions have
multiple components, studies should follow the
guidelines for complex intervention development
[69]. Key questions are as follows:
• At what level does the intervention work (e.g.
patient-level, practice-level, hospital level)?
• How long does the intervention last, or is it a
‘one-off’?
• Who is delivering the intervention (e.g. a practitioner or a researcher)?
• Is the delivery of the intervention standardised?
3.3 The Comparator
Of similar importance to the intervention being
evaluated is the denition, description, and integrity of the comparator as this will also affect both
effectiveness and cost aspects of the evaluation.
Many studies use ‘current practice’ as the com-

5 Economics ofMedication Safety, withaFocus onPreventable Harm
119
parator, but it is not always clear what this is. As
such, current practice may differ between settings making comparison or extrapolation of
study results difcult.
Some studies do not collect primary data for
the comparator, but rather model the counterfactual by making assumptions about (or consulting
expert opinion on) what would have happened to
the patient [70, 71] or resource consumption [72]
in the absence of the intervention. For example,
Dooley etal. took ward-based interventions made
by pharmacists in eight Australian hospitals and
asked an expert panel to estimate the importance
of the intervention and any associated treatment
and outcomes that may have been averted by that
intervention [71]. They then attached unit costs to
those data to estimate the costs of the events
averted by the intervention. This method is sometimes used in modelling when primary data are
not available but presents considerable uncertainty around an effect size.
Given that many interventions are based
around pharmacists identifying problems with
medicines that have been created or missed by
others, the current practice comparator is sometimes non-disclosure of this information to the
relevant party in a primary care [73, 74] or secondary care [75] setting. A study of prescriber
support described a comparator group where the
‘pharmacist proposed recommendations that
remained concealed from the physicians’ and
the physicians were told to ‘act as though the
pharmacist present in the ofce is invisible’
[74]. However, withholding information that
could affect patient outcome is usually perceived to be unethical in healthcare settings
where equipoise no longer exists regarding the
effect of the intervention. To handle this, some
studies have derived comparators where the
‘control’ was the pharmacist providing written
information and the intervention was one of a
range of face-to-face interactions between the
pharmacist and prescriber [76, 77]. The adaptation of this comparator clearly represents a different evaluation when compared with a
comparator that has no input from the pharmacist at all, the latter approach is likely to show a
bigger effect of the intervention. For example,
Bond etal. compare standard practice with the
intervention of giving written medication review
feedback to GPs [73]. Interestingly, this intervention was equivalent to the control arm used
in other studies that compared written feedback
on prescribing with face-to-face meetings or
case conferences [76, 77].
3.4 Study Design
An economic evaluation requires a comparative
study to provide an estimate of relative effectiveness when comparing one intervention with
another, or with current practice. Therefore, economic evaluations are often carried out alongside
RCTs. RCTs are considered the gold standard
source for effectiveness evidence. It can be a
challenge to design and carry out RCTs for complex safety interventions that may function at a
practice or hospital level. Contamination between
intervention and control arms can be dealt with
by clustering at the practice or hospital level [76,
78]. In complex safety interventions, it should be
recognised that it is not always possible to carry
out an RCT, and therefore other non-randomised
comparative study designs can be used. These
may use routinely collected data or observational
data collected specically for the study. Study
designs tend to be either historical or concurrent
controls or both. This type of study will have the
possibility of contamination or secular changes,
as well leading to the introduction of selection
bias through the non-random allocation of intervention and control sites. Estimation of effect
sizes in these types of studies need to take account
of these potential biases in the data.
A full economic evaluation requires presentation of the costs and outcomes of the intervention
and its comparator to allow assessment of the
incremental cost and benet. Some published
studies present partial economic evaluations
where no incremental analysis is presented, or
cost–consequences analyses, which examine
costs and consequences without attempting to
isolate a single consequence or aggregate consequences into a single measure with no incremental analysis [51]. Budget impact analysis is
Соседние файлы в папке Библиотека им академика М.И. Перельмана
