health

1. Model-based cost-effectiveness analysis of interventions
aimed at preventing medication error at hospital admission
(medicines reconciliation)
Jonathan Karnon MSc PhD,1
Fiona Campbell RGN2
and Carolyn Czoski-Murray RGN MSc2
1
Associate Professor, 2
Research Fellow, School of Health and Related Research, University of Sheffield, Sheffield, UK
Keywords
cost-effectiveness, decision analysis,
medication errors, medicines reconciliation,
probabilistic calibration, QALYs
Correspondence
Jonathan Karnon
School of Population Health and Clinical
Practice
Level 9, 10 Pulteney Street
Maildrop 207
University of Adelaide
Adelaide 5005
South Australia
Australia
E-mail: jonathan.karnon@adelaide.edu.au
Accepted for publication: 28 January 2008
doi:10.1111/j.1365-2753.2008.01000.x
Abstract
Rationale Medication errors can lead to preventable adverse drug events (pADEs) that
have significant cost and health implications. Errors often occur at care interfaces, and
various interventions have been devised to reduce medication errors at the point of admis-
sion to hospital. The aim of this study is to assess the incremental costs and effects
[measured as quality adjusted life years (QALYs)] of a range of such interventions for
which evidence of effectiveness exists.
Methods A previously published medication errors model was adapted to describe the
pathway of errors occurring at admission through to the occurrence of pADEs. The baseline
model was populated using literature-based values, and then calibrated to observed outputs.
Evidence of effects was derived from a systematic review of interventions aimed at
preventing medication error at hospital admission.
Results All five interventions, for which evidence of effectiveness was identified, are
estimated to be extremely cost-effective when compared with the baseline scenario.
Pharmacist-led reconciliation intervention has the highest expected net benefits, and a
probability of being cost-effective of over 60% by a QALY value of £10 000.
Conclusions The medication errors model provides reasonably strong evidence that some
form of intervention to improve medicines reconciliation is a cost-effective use of NHS
resources. The variation in the reported effectiveness of the few identified studies of
medication error interventions illustrates the need for extreme attention to detail in the
development of interventions, but also in their evaluation and may justify the primary
evaluation of more than one specification of included interventions.
Introduction
Medication error is a leading cause of avoidable harm suffered by
patients, resulting in significantly increased morbidity, prolonged
length of stay in hospital and increased mortality [1]. Medication
error occurs most commonly at the interfaces of care [2]. At point
of hospital admission, variances between the medications patients
were taking prior to admission and their prescriptions on admis-
sion ranged from 30% to 70% in two recent literature reviews
[3,4]. The need for improvements in medicines management
arrangements at admission were highlighted by the Audit Com-
mission report, ‘A Spoonful of Sugar’ [5] and the NPSA report,
‘Moving patients medicines safely’ [6].
Medicines reconciliation is defined as ‘being the process of
identifying the most accurate list of a patient’s current medicines –
including the name, dosage, frequency and route – and comparing
them to the current list in use, recognizing any discrepancies, and
documenting any changes, thus resulting in a complete list of
medications, accurately communicated’ [7]. The aim of this study
was to address the evidence around interventions aimed at reduc-
ing errors in the medicines reconciliation process, and to estimate
the incremental costs and quality adjusted life years (QALYs) of
such interventions.
Model structure
The economic analysis used an adapted version of a previously
developed medication errors model [8], which covered three
phases of hospital care at which medication errors could occur:
prescribing, dispensing and administration. Subsequent to the
occurrence of an error, the error could be detected prior to reaching
the patient. If the error reached the patient, it was assigned a
probability of causing harm, and categorized as having minor,
moderate or severe health effects. The adapted model (shown in
Journal of Evaluation in Clinical Practice ISSN 1356-1294
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd, Journal of Evaluation in Clinical Practice 15 (2009) 299–306 299

2. Fig. 1) is specific to errors occurring in the medicines reconcilia-
tion process. The choice of error types included was based on an
analysis of the most common and/or severe error types occurring at
the reconciliation stage. The literature indicates that few pADEs
occur as a result of omitted drugs (i.e. drugs patients were receiv-
ing before admission that are not prescribed in hospital) [9–13].
However, as these are one of the most frequently cited error types
occurring at the medicines reconciliation stage, drug omission is
included as a category of error. The other defined error types
include errors of commission and known allergy errors [9,14].
Commission errors were combined as few studies presented sepa-
rate commission error rates, and there is very little evidence to
inform separate estimates of detection, and probability of harm.
Model population
Input parameter values for the model are presented in Tables 1 and
2. The following section describes the corresponding data sources;
further details are available in the technical report [15].
In estimating the frequency of medicines reconciliation errors,
it is important to distinguish between errors in the obtained
drug histories and prescription errors – a drug history error
may not lead to a prescription error as it may be rectified via
alternative information sources. The estimation of the baseline
rate of prescription errors per patient occurring because of defi-
ciencies in medicines reconciliation (including notification of
drug allergies) was informed by the control arm of the only UK
intervention study that included an independent assessment of
errors [16]. The rates were converted to rates per prescription
order based on a recent audit undertaken at the Royal Hallam-
shire Hospital in Sheffield (Personal communication: Nicky
Thomas, Sheffield Teaching Hospitals NHS Foundation Trust.
2007), from which a mean number of six (range four to eight)
prescription orders were estimated to be received following
admission.
Five studies were identified that presented a wide range of error
detection rates, because of differences in the definition of unde-
tected errors and methods of data collection. As the model
describes the pathway of all errors (not just those with the potential
to cause harm), it is likely that the relevant detection rates
are lower than the reported detection rates for potential pADEs.
[10,17] The estimated detection rates were informed by three
studies that report aggregate rates [18–20], with adjustments for
the higher detection rate for prescription errors reported by Scarsi
et al. [20] as well as the calibrated detection rates from the previ-
ous medication errors model [8]. Differentiation between the error
types is informed by the relative likelihood of pADEs being caused
by the different error types (as described below).
No error
Error detected prior to reaching patient
Minor harm
Moderate harm
Severe harm
Error causes harm
Error causes no harm
Error not detected
Error of omission
Error of commission
Allergy not recorded
Patient admitted
Figure 1 Medicines reconciliation error model structure.
Table 1 Mean and 95% confidence intervals
or ranges for calibrated input parameters
Input parameters Pre-calibration Post-calibration
Error probabilities per prescription order
Errors of omission 0.400 (0.300–0.590) 0.364 (0.302–0.551)
Errors of commission 0.160 (0.120–0.240) 0.173 (0.123–0.234)
Errors because of known allergies 0.070 (0.050–0.110) 0.081 (0.051–0.109)
Prescription error detection probabilities
Errors of omission 40–70% 0.617 (0.433–0.699)
Errors of commission 20–50% 0.379 (0.224–0.490)
Errors because of known allergies 40–70% 0.558 (0.415–0.693)
Probabilities of harm for undetected errors
Errors of omission 0.1–1% 0.001 (0.001–0.002)
Errors of commission 1–5% 0.020 (0.011–0.036)
Errors because of known allergies 0.1–1% 0.005 (0.001–0.010)
Economic analysis of medicines reconciliation J. Karnon et al.
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd
300

3. Wilson et al. [18] showed that nine of 143 (6.3%) undetected
errors were classified as requiring increased monitoring or worse,
although this study relied on error reports and so requires the
assumption that probabilities of harm are similar in reported and
unreported errors. Kaushal et al. reported that 9.6% of non-
intercepted errors with the potential to cause harm actually caused
harm in a paediatric unit [17]; the corresponding figure reported by
Bates et al. for all units was 38.5% [10]. The estimated ranges for
the probabilities of harm for all errors were set lower.
Estimates of the distribution of severity for pADEs were taken
from two relevant sources [10,21], for which the weighted average
proportions were 20% fatal/life-threatening (resulted in permanent
patient harm, required intervention to sustain life, or contributed to
a patient’s death); 41% serious (resulted in temporary harm to the
patient and required prolonged hospitalization); 39% significant
(resulted in temporary harm to the patient and required interven-
tion without increased length of stay).
Calibration
To calibrate the model, outputs for 10 000 sets of input parameter
values were compared with estimates of the pADE rate by error
type.
Six studies were identified that reported aggregate rates across
error types and medication stages for pADEs [10,17,21–24]. After
adjustment to obtain error rates per 1000 prescription orders, and
exclusion of an apparent outlier study [24], the range of aggregate
pADE rates was estimated to be between one and four per 1000
orders. Studies show that most pADEs originate at the ordering
and administration stages [17,22,23]. The proportion of events
originating as ordering errors ranges from 0.41 for preventable and
potential (intercepted & non-intercepted) ADEs in ICU to 0.91 for
potential pADEs in paediatrics. Lower and upper bounds of the
proportion of pADEs originating at the prescription stage were
defined as 0.5 and 0.85 respectively. Thus, the range for the rate of
pADEs originating at the prescription stage was between 0.5 and
3.4 per 1000 orders.
Data on the median proportion of pADEs by error type informed
ranges for each of the error types [14]. One quarter of the ‘other’
error type category was assumed to represent ‘errors of omission’,
as these errors were not explicitly represented (the other ‘other’
errors were assumed to be errors of commission). To obtain joint
distributions, values were randomly sampled from uniform distri-
butions for each error type, which were proportionally adjusted to
sum to 100.
The first stage of the calibration process involved identifying
eligible input parameter sets. Eligibility was defined as sets that
predicted aggregate pADE rates (for medicines reconciliation-
related prescription errors) that fell within the estimated observed
95% confidence interval (CI). Within the eligible set, the following
steps involved:
• summing the distances for each of the three output parameters to
estimate the aggregate absolute difference;
• defining the reciprocal of the aggregate difference (one divided
by the difference) as the weight for each input parameter set that
reflects how closely each set predicts the observed output param-
eter values; and
• defining probabilities that each input parameter set was the
optimal set as the estimated weight for each parameter set divided
by the sum of the weights across all eligible sets.
Intervention effectiveness
Intervention effectiveness was described as the relative risk (RR)
of medication errors occurring with an intervention in place com-
pared to the baseline scenario, which then feeds through the model
to estimate the corresponding reduction in pADEs.
Table 2 Mean and 95% confidence intervals
or ranges for non-calibrated input parameters
Parameters Model values
Relative risks (RRs)
Pharmacist-led reconciliation 0.250 (0.100–0.400)
Standardized forms, pharmacy technicians, hospital policy 0.480 (0.330–0.630)
Nurses taking histories with standardized form 0.375 (0.225–0.525)
Computerized assessment and feedback by pharmacist 0.430 (0.280–0.580)
Current medication faxed from the GP practice 0.310 (0.160–0.460)
Intervention costs
Pharmacist-led reconciliation 10.280 (5.580–21.390)
Standardized forms, pharmacy technicians, hospital policy 8.500 (4.690–17.500)
Nurses taking histories with standardized form 14.710 (6.020–38.150)
Computerized assessment and feedback by pharmacist 14.505 (9.150–19.840)
Current medication faxed from the GP practice 9.260 (5.780–12.730)
Cost parameter
Detected medication errors £0–£6
Significant (non-increased LoS) pADEs length of stay £65–£150
Serious pADEs £713–£1484
Severe, life-threatening, or fatal pADEs £1085–£2120
QALY loss by pADE severity
Significant 0.001–0.008
Serious 0.061–0.09
Severe/life-threatening/fatal (full range 0.7–12.8) 1–4.41
LoS, length of stay; pADEs, preventable adverse drug events; QALY, quality adjusted life years.
J. Karnon et al. Economic analysis of medicines reconciliation
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd 301

4. Eleven databases, the Internet, and electronic discussion lists
were searched up to July 2007 for evidence on the effectiveness of
interventions that sought to achieve medicines reconciliation. Key
terms used to conduct the search were identified in consultation
with clinical experts. Given the lack of randomized controlled
trials, non-randomized studies were included in the review. Studies
were excluded if the method of comparison was non-randomized
with a non-independent assessment of outcomes, that is, where the
intervention was treated as the gold standard. Inclusion criteria
specified that only studies with outcomes defined as either pADEs,
potential pADEs, or medication errors were included.
Evidence on three broad categories of intervention aimed at
preventing errors during the medicines reconciliation process was
identified: pharmacist involvement in the medicines reconciliation
process; medicines reconciliation package involving the develop-
ment of a standardized medication form; and IT based/Information
transfer initiatives. However, variations in methods, settings, and
definitions precluded any meaningful combination of the results
of studies evaluating similar interventions, so subjectively defined
ranges of the RRs were defined. The defined RRs were applied to
all error types as it was not considered feasible to differentiate
between different error types. It is also necessarily assumed that
the relationship between medication errors and pADEs is constant,
that is, a reduction in the incidence of medication errors leads to a
proportional reduction in the incidence of pADEs.
Two UK studies evaluated the effectiveness of pharmacist
involvement in the medicines reconciliation process using an inde-
pendent source to verify errors [20,22]. Collins reported little
difference between pharmacist and physician aggregate effective-
ness [25], although this finding was confounded by the cases in
which the pharmacist had been unable to obtain histories.
McFadzean found a higher error rate in physicians (5 vs. 65 errors
per 100 patients) [16], while drug allergies were recorded in 23%
and 93% of physician and pharmacist-based charts. The only ran-
domized controlled trial (RCT) compared a preoperative struc-
tured pharmacist medication assessment with standard care in
Canada [26]. The pharmacist assessment reduced the number of
patients with at least one medication discrepancy from 43.6% to
19.5%, a RR of 0.45.
Three US-based studies evaluating medicines reconciliation
packages were reviewed. A combination of standardized forms,
pharmacy technicians, and hospital policy initiatives to improve
the quality of orders reduced defects per drug order from 0.25 to
0.12, a RR of 0.48 [27]. The use of nurses to take medication
histories aided by a standardized form decreased errors per 100
admissions from 213 to 80, a RR of 0.375 [28]. Computerized
assessment and feedback to physicians of patients’ medication
profiles by a pharmacist achieved a RR of 0.43 for patients with
order discrepancies [29].
A UK-based before and after study evaluated the transfer of
current medication and other relevant information by Fax: from
the GP practice to the admitting ward [30]. The Fax: system
reduced errors from 55 per 100 patients to 17, a RR of 0.31.
Intervention implementation costs
The main cost of a system of pharmacist-led medicines reconcili-
ation is assumed to be the additional time requirements for the
pharmacists. Dutton et al. report an increase in the mean time
spent on a ward per day of 81.6 minutes following the introduction
of pharmacists taking medication histories [31]. Based on an
average ward capacity of 30 beds and an average length of stay
of 8.1 days [32], 3.7 new patients would be expected each day.
The mean additional time per patient receiving pharmacist-led
medicines reconciliation is 81.6/3.7 = 22 minutes (95% CI
12–46 minutes, assuming a 95% CI of 2.2–5.2 for daily admis-
sions). An hourly cost of a pharmacist of £28 was based on the
mid-point of Agenda for Change (AfC) salaries band 6 of the April
2005 pay scale [33].
Time inputs for pharmacists to the medicines reconciliation
packages were assumed to be similar to those estimated above,
although more junior staff were assumed (AfC salaries band 5).
Other elements are less tangible. It was assumed that the develop-
ment and maintenance of the standardized form requires the
equivalent of 1 week’s work of a pharmacist (£1050) each year,
with the assigned development and maintenance cost as £0.06
per admission. Costs of dissemination of a new hospital policy
were assumed to require 15 minutes of every prescriber’s time,
two-thirds of which were assigned to the interventions effects on
improved medicines reconciliation.
Nursing time to take histories with a standardized form was
estimated by applying the 81% increase in time required to obtain
medication histories identified by Nester [34], to the estimated
pharmacist time (22 minutes). The AfC salaries mid-point for band
5 was applied to the time estimates. The same cost per admission of
£0.06 to develop and maintain the standardized form is assumed.
Computerized assessment and feedback by a pharmacist to
inform doctor-led medicines reconciliation was assumed to require
an additional 11 minutes of pharmacists’ time (half the amount
required to take a medication history), and an additional 5 minutes
of prescribing physicians’ time. The upper bound of the cost of
setting up a computerized system is based on the lower cost bound
associated with setting up and maintaining a Computerised Phy-
sician Order Entry (CPOE) system [35]. Lower bounds were
specified as one quarter of the upper bounds (based on the
observed range for CPOE costs). Set-up costs were annuitized at
3.5% per annum assuming a 10-year useful life for the system.
A system of faxed current medication lists from patients’ GP
assumed similar development costs to those estimated above, and
that a member of the practice’s clerical staff could complete the
form and submit it, requiring a mean time of 10 minutes (AfC
salaries mid-point for band 2 is assumed clerical staff). The use of
the form by the prescribing physician is assumed to add
7.5 minutes.
Interventions may have some additional cost savings related to
reductions in ancillary test usage (mostly laboratory) and reduc-
tions in length of stay via guideline embedding and variance analy-
sis. However, these potential savings are not included because of a
lack of evidence.
Costs of pADEs
All of the identified data describing additional treatment costs for
patients experiencing an adverse drug event are US-based. Bates
et al. undertook a case control costing study that defined two sets
of cases as patients with an ADE, and patients with a pADE [36].
Controls were selected as patients on the same unit as the case with
the most similar pre-event length of stay (LoS). Differences were
Economic analysis of medicines reconciliation J. Karnon et al.
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd
302

5. greatest for patients with pADEs compared with controls: length
of stay was 4.6 days longer for patients (P = 0.03), total charges
were $11 524 higher for patients (P = 0.06), and total costs were
$5857 higher for patients (P = 0.07).
Other case control studies report additional costs of all ADEs
and adverse events (including falls and surgical mishaps) of $2262
and $2,411 respectively [37,38]. Based on a retrospective chart
review, Schneider et al. estimated the cost of medication errors
requiring extra laboratory tests or treatment without an increased
LoS to be $95 to $227, of errors resulting in a prolonged length of
stay to be $2596, and of errors resulting in near-death experience
to be $2640 [39].
The Leapfrog Group in the US reported that medication errors
cost $10 [40], which was used as an upper bound for errors that do
not lead to an ADE.
QALY effects
QALYs losses because of pADEs were estimated using two dis-
tinct methods. A dataset of financial claims made against the NHS,
including 251 non-zero closed cases involving medication errors,
was obtained from the NHS Litigation Authority. The payouts
ranged from £17 to over £0.5 million. The database provided
intermittent descriptions of the outcomes making it difficult to link
injuries to the model-defined categories. Therefore, payments
were arranged in order and approximate ranges for the significant,
serious, and severe/life-threatening/fatal categories were assigned
as the first to twentieth, thirtieth to sixtieth, and seventieth to
ninety-ninth percentiles respectively. The National Institute for
Clinical Excellence (NICE) has been described as applying a value
of between £20 000 and £30 000 per QALY gained [23], which
was applied to the monetary values to estimate QALY losses.
No relevant data describing the utility effects of the broadly
defined severity categories were identified, and so hypothetical
estimates were based on assumed utility decrements for each cat-
egory and an accompanying duration of effect. The utility decre-
ment describes the reduction in the quality of life of a patient as a
result of a pADE, a utility decrement of 0.1 indicates an absolute
reduction of 10% on a scale of 0–100. Table 3 describes the
assumptions for each of the three pADE categories, which were
based on discussions within the research team and a retrospective
study that estimated that 43% (95% CI, 35%-51%) of patients
who died following an error defined as definitely or probably
preventable would have left the hospital alive given optimal care
[41]. These cases inform the lower bound QALY loss for severe
ADEs.
The estimated QALY losses for significant pADEs are small and
similar between the two methods and the full range of uncertainty
is incorporated. The other categories show more variation. As the
model requires estimates of the mean QALY loss across all pADEs
within each category, the extreme values are discarded from the
four presented estimates for each category, and the middle values
used in the model.
Model analysis
The model was analysed by sampling 10 000 input parameter
sets based on the probability that they represent the optimal set.
Additional parameter values were sampled from probability dis-
tributions representing severity of incident pADEs, intervention
effectiveness, implementation costs, and pADE cost and QALYs
effects. The RRs and cost parameters were represented as log
normal distributions: bounded at zero with a long tail representing
the small likelihood of limited and even negative effectiveness or
large costs respectively.
Outputs were analysed to estimate the mean incremental cost
per QALY gained of each intervention compared with the baseline
scenario, as well as a cost-effectiveness acceptability frontier.
Frontiers describe the probability that the intervention with the
highest expected net benefits at alternative QALY values is the
most cost-effective intervention (estimated as the proportion of
the 10 000 iterations in which that intervention has the highest net
benefits). Net benefits equal QALYs gained multiplied by the
assumed value of a QALY minus costs.
Results
Given the input parameter ranges specified in the above sections,
the calibration process identified 2328 eligible input parameter
sets from the 10 000 sets that were analysed. The resulting differ-
ence between the pre- and post-calibration input parameter values
are presented in Table 1.
The main outputs from the model are described in Table 4,
which show the costs and numbers of non-intercepted medication
errors and pADEs occurring with every 1000 prescription orders,
and the corresponding loss of QALYs. The results show that the
nurse-based reconciliation intervention has the highest interven-
tion costs, because of the observation that nurses take considerably
longer than pharmacists to take a medication history. The comput-
erized assessment approach has the second largest cost, because of
the assumed cost of setting up such a system.
In terms of effectiveness, pharmacist-led reconciliation is pre-
dicted to prevent the most medication errors, followed by a system
involving faxed details from a patient’s General Practice. The
health gains show that the prevention of one pADE corresponds to
a gain of approximately one QALY, and that the largest QALY gain
is 2.2 QALYs per 1000 orders. Substantial costs savings because
of the prevention of pADEs are also predicted.
Table 3 Assumed QALY-based monetary valuations of the pADE sever-
ity categories
Significant: resulted in temporary harm to the patient and required
intervention
Utility decrement 0.1 0.2
Effect duration 3 days 14 days
Serious: resulted in temporary harm to the patient and required initial
or prolonged hospitalization
Utility decrement 0.2 0.4
Effect duration 14 days 56 days
Severe, life-threatening, or fatal: resulted in permanent patient harm,
required intervention to sustain life or contributed to a patient’s
death.
Utility decrement 1 0.3
Effect duration 1 years 20 years
PADEs, preventable adverse drug events; QALY, quality adjusted life
years.
J. Karnon et al. Economic analysis of medicines reconciliation
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd 303

6. The incremental cost per QALY gained (ICQ) results show that
all five interventions are estimated to be extremely cost-effective
when compared with the baseline scenario. Three of the interven-
tions are shown to dominate the baseline scenario (i.e. cost less and
gain more), while the upper CI for all five interventions is below
£5000.
Figure 2 presents the cost-effectiveness acceptability frontier,
which shows that the ‘GP fax’ intervention has the highest mean
net benefits when the QALY value is zero, but that as soon as any
value is attached to a QALY gain the pharmacist-led reconciliation
intervention becomes the preferred intervention. The probability
that this intervention is cost-effective rises to over 60% by a QALY
value of £10 000, at which point it plateaus.
Discussion
This study assessed the cost-effectiveness of interventions aimed
at preventing medication errors occurring at the point of admission
to hospital, through improvements in reconciling medicines
received prior to admission and medicines received in-hospital.
Five separate interventions for which some effectiveness evi-
dence was identified were evaluated with respect to a baseline
scenario. The presented results show that a pharmacist-led medi-
cines reconciliation intervention is likely to provide the largest net
benefits to the NHS. This result is observed despite the conserva-
tive assumption that the additional employment of pharmacists to
assist in the medicines reconciliation process will not free up
physicians time to be spent on other activities that would provide
health benefits. In the absence of pharmacist capacity, it may be
more feasible to implement a system of faxed medication sheets
from patients’ General Practices.
Limitations include the reliability of data to inform model input
parameters, such as reliance on US data. It was difficult to adjust
data to a UK context as the overall direction of the many differ-
ences between the US and the UK, in terms of increasing or
decreasing the aggregate rate of pADEs, is unclear. The process of
defining feasible input parameter ranges, and calibrating to esti-
mated output parameter values provides some assurance for the
included parameters. There is more uncertainty around the non-
calibrated parameters, especially severity of pADEs, which was
Table
4
Model
outputs:
mean
values
(95%
confidence
intervals)
per
1000
prescription
orders
Intervention
Intervention
costs
Error
costs
Total
costs
Non-intercepted
medication
errors
PADEs
Total
QALYs
lost
Inc.
cost
per
QALY
gained
Baseline
£0
(£0–£0)
£4092
(£2072–£6758)
£4092
(£2072–£6758)
323
(215–456)
2.8
(1.5–4.5)
3.0
(0.9–7.0)
Dominates
(Dominates-£1177)
Pharmacist-led
reconciliation
£1897
(£811–£3785)
£1090
(£390–£2362)
£2987
(£1565–£5229)
86
(36–170)
0.7
(0.3–1.6)
0.8
(0.2–2.2)
Dominates
(Dominates-£1695)
Standardized
forms,
pharmacy
technicians,
hospital
policy
£1552
(£689–£3059)
£1990
(£922–£3538)
£3543
(£2029–£5632)
157
(93–243)
1.4
(0.7–2.4)
1.5
(0.4–3.6)
£184
(Dominates-£4402)
Nurses
taking
histories
with
standardized
form
£2866
(£897–£6868)
£1567
(£697–£2938)
£4433
(£2106–£8525)
124
(68–205)
1.1
(0.5–2.0)
1.1
(0.3–2.9)
£138
(Dominates-£3124)
Computerized
assessment
and
feedback
by
pharmacist
£2542
(£1469–£4230)
£1783
(£822–£3222)
£4325
(£2752–£6445)
141
(81–225)
1.2
(0.6–2.2)
1.3
(0.3–3.1)
Dominates
(Dominates-£623)
Current
medication
faxed
from
the
GP
practice
£1632
(£923–£2737)
£1314
(£542–£2596)
£2945
(£1816–£4588)
104
(52–184)
0.9
(0.4–1.8)
1.0
(0.2–2.5)
Dominates
(Dominates-£1177)
PADEs,
preventable
adverse
drug
events;
QALY,
quality
adjusted
life
years.
0
0.2
0.4
0.6
0.8
1
0
Value of a QALY (£000s)
Probability
intervention
with
highest
mean
NBs
is
cost-effective
Faxed form from GP
Pharmacist-led reconciliation
5 10 15 20 25 30 35 40 45 50
Figure 2 Cost-effectiveness acceptability frontier for interventions
aimed at improving medicines reconciliation. QALY, quality adjusted life
years.
Economic analysis of medicines reconciliation J. Karnon et al.
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd
304

7. based solely on a couple of related US studies [10,21]. Until the
validity of incident reporting is assured (i.e. that incident reporting
identifies a complete and unbiased set of ADEs), relevant UK
studies are required to inform these parameters. Similarly, quality
of life studies relating to the effects of pADEs would usefully
inform the model.
The assumption of proportionality between the prevention of
medication errors and the occurrence of pADEs is difficult to
justify. This is because there is such a wide range of potential
medication errors with different potential health impacts, as well
as different likelihoods of detection. It is likely that interventions
prevent alternative types of errors differentially. Bates [42] provide
an extreme example, in which the RR for all medication errors for
a CPOE system compared with baseline (in the first implementa-
tion period) was 0.37 (145.2 vs. 53.6 errors per 1000 patient days),
while the RR for pADEs was 1.97 (2.9 vs. 5.7 per 1000 patient
days). Prospective studies that investigate the relationship between
medication errors and ADEs from a UK perspective would help,
although there are many types of medication errors, each of which
will have different probabilities of detection prior to administra-
tion, of causing harm, and of causing different levels of severity of
harm.
The precise specification of alternative interventions is likely to
alter their effectiveness, including factors such as rates of clinician
acceptance and ease of use. A recent, primarily qualitative study
concluded that as systems are implemented, clinicians and hospi-
tals must try to minimize errors that these systems cause in addi-
tion to errors that they prevent [43]. In the context of the current
study, for example, this may involve setting explicit ground rules
for interactions between pharmacists and clinicians while jointly
attending ward rounds.
Conclusions
The medication errors model provides reasonably strong evidence
that some form of intervention to improve medicines reconcilia-
tion is a cost-effective use of NHS resources. The results indicate
that pharmacist-led medicines reconciliation is likely to be the
most cost-effective intervention, although it is difficult to assess
whether the model has captured all of the relevant uncertainty.
There are also likely to be other interventions, particularly
IT-based interventions, for which evidence of effectiveness was
not available.
The variation in the reported effectiveness of the few identified
studies of medication error interventions illustrates the need for
extreme attention to detail in the development of interventions, but
also in their evaluation and may justify the evaluation of more than
one specification of included interventions. Key drivers of cost-
effectiveness should be specifically addressed in the design of
evaluations of medication error interventions, in particular, data
should be collected on the severity of ADEs occurring in the
different intervention groups.
If further research confirms the cost-effectiveness of
pharmacist-led medicines reconciliation, the capacity of the NHS
to employ more pharmacists will be a key factor in the implemen-
tation of this intervention. Solutions to the supply issue should be
considered at the same time as the evaluation of the intervention as
some solutions may affect the design of evaluation studies. Critical
incidence studies may be undertaken to define the attributes of
pharmacists that contribute most to the reduction of medication
errors, which may identify interventions such as new training
programmes for other health professionals and new processes of
health care delivery.
References
1. Pirmohamed, M., James, S., Meakin, S., Green, C., Scott, A. K.,
Walley, T. J., Farrar, K., Park, B. K. & Breckenridge, A. M. (2004)
Adverse drug reactions as cause of admission to hospital: prospective
analysis of 18 820 patients. British Medical Journ, 329, 15–19.
2. Rozich, J. D. & Resar, R. K. (2001) Medication safety: one organiza-
tion’s approach to the challenge. Journal of Clinical Outcomes Man-
agement, 8 (10), 27–34.
3. Cornish, P. L., Knowles, S. R., Marchesano, R., Tam, V., Shadowitz,
S., Juurlink, D. N. & Etchells, E. E. (2005) Unintended medication
discrepancies at the time of hospital admission. Archives of Interna-
tional Medicine, 165, 424–429.
4. Gleason, K. M., Groszek, J. M., Sullivan, C., Rooney, D., Barnard, C.
& Noskin, G. A. (2004) Reconciliation of discrepancies in medication
histories and admission orders of newly hospitalized patients. Ameri-
can Journal of Health-System Pharmacy, 61, 1689–1695.
5. Audit Commission (2001) A spoonful of sugar – medicines manage-
ment in hospitals. London: Audit Commission. Available from http://
www.audit-commission.gov.uk/Products/NATIONAL-REPORT/
E83C8921-6CEA-4b2c-83E7-F80954A80F85/nrspoonfulsugar.pdf
(accessed 24 February 2009)
6. The Royal Pharmaceutical Society of Great Britain (2006)
Moving patients medicines safely. Guidance on Discharge and
Transfer Planning. Available from http://www.rpsgb.org.uk/pdfs/
dischtransfplanwbook.pdf (accessed 24 February 2009)
7. Manno, M. S. & Hayes, D. D. (2006) How medication reconciliation
saves lives. Nursing, 36 (3), 63–64.
8. Karnon, J., McIntosh, A., Dean, J. et al. (2007) A prospective hazard
and improvement analytic approach to predicting the effectiveness of
medication error interventions. Safety Science, 45, 523–539.
9. Winterstein, A. G., Hatton, R. C., Gonzalez-Rothi, R., Johns, T. E. &
Segal, R. (2002) Identifying clinically significant preventable adverse
drug events through a hospital’s database of adverse drug reaction
reports. American Journal of Health-Systematic Pharm, 59, 1742–
1749.
10. Bates, D. W., Boyle, D. L., Vliet, M. V. V. & Schneider, J. & Leape,
L. (1995) Relationship between medication errors and adverse drug
events. Journal of International Medicine, 10, 199–205.
11. Kaushal, R., Bates, D. W., Landrigan, C., McKenna, K. J., Clapp, M.
D., Federico, F. & Goldmann, D. A. (2001) Medication errors and
adverse drug events in pediatric inpatients. Journal of the American
Medical Association, 285, 2114–2120.
12. Lesar, T. S., Briceland, L. L., Delcoure, K., Parmalee, J. C., Masta-
Gornic, V. & Pohl, H. (1990) Medication prescribing errors in a
teaching hospital. Journal of the American Medical Association, 263,
2329–2334.
13. Bobb, A., Gleason, K., Husch, M., Feinglass, J., Yarnold, P. R. &
Noskin, G. A. (2004) The epidemiology of prescribing errors: the
potential impact of computerized prescriber order entry. Archives of
International Medicine, 164, 785–792.
14. Kanjanarat, P., Winterstein, A. G., Johns, T. E., Hatton, R. C.,
Gonzalez-Rothi, R. & Segal, R. (2003) Nature of preventable adverse
drug events in hospitals. A Literature Review, 60, 1750–1759.
15. Campbell, F., Karnon, J., Czoski-Murray, C. & Jones, R. (2007) A
systematic review of the effectiveness and cost-effectiveness of inter-
ventions aimed at preventing medication error (medicines reconcilia-
tion) at hospital admission. Report for the National Institute for Health
and Clinical Excellence as part of the Patient Safety Pilot.
J. Karnon et al. Economic analysis of medicines reconciliation
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd 305

8. 16. Mcfazdean, E., Isles, C., Moffat, J., Norrie, J. & Steward, D. I. (2003)
Is there a role for a prescribing pharmacist in preventing prescribing
errors in the medical admissions ward? Pharmaceutical Journal, 270,
896–899.
17. Kaushal, R., Bates, D. W., Landrigan, C., McKenna, D. J., Clapp, M.
D. & Federico, F. (2001) Medication errors and adverse drug events in
pediatric inpatients. Journal of the American Medical Association,
285, 2114–2120.
18. Wilson, D. G., McArtney, R. G., Newcombe, R. G., McCartney, R. J.,
Gracie, J. & Kirk, C. R. (1998) Medication errors in paediatric prac-
tice: insights from a continuous quality improvement approach. Euro-
pean Journal of Pediatrics, 157, 769–774.
19. Dale, A., Copeland, R. & Barton, R. (2007) Prescribing errors on
medical wards and the impact of clinical pharmacists. International
Journal of Pharmacy Practice, 11, 19–24.
20. Scarsi, K. K., Fotis, M. A. & Noskin, G. A. (2002) Pharmacist par-
ticipation in medical rounds reduces medication errors. American
Journal of Health-System Pharmacy, 59, 2089–2092.
21. Bates, D. W., Cullen, D. J., Laird, N. et al. (1995) Incidence of adverse
drug events and potential adverse drug events – implications for pre-
vention. Journal of the American Medical Association, 274, 29–34.
22. Bates, D. W., Leape, L. L. & Petrycki, S. (1993) Incidence and
preventability of adverse drug events in hospitalized adults. Journal of
General International Medicine, 8, 289–294.
23. Cullen, D. J., Sweitzer, B. J., Bates, D. W., Burdick, E., Edmondson,
A. & Leape, L. L. (1997) Preventable adverse drug events in hospi-
talized patients: a comparative study of intensive care and general care
units. Critical Care Medicine, 25, 1289–1297.
24. Bond, C. A., Raehl, C. L. & Pitterle, M. E. (1994) 1992 National
Clinical Pharmacy Services Study. Pharmacotherapy, 14, 282–304.
25. Collins, D. J., Nickless, G. D. & Green, C. F. (2004) Medication
histories: does anyone know what medicines a patient should be
taking? International Journal of Pharmacy Practice, 12, 173–178.
26. Kwan, Y., Fernandes, O., Nagge, J., Wong, G., Huh, J., Hurn, D. &
Bajcar, J. (2005) Implementation and a randomized controlled
evaluation of pharmacist medication assessments in a surgical pre-
admission clinic. Pharmacotherapy, 25, 1462–1462.
27. Michels, R. D. & Meisel, S. B. (2003) Program using pharmacy
technicians to obtain medication histories. American Journal of
Health-System Pharmacy, 60, 1982–1986.
28. Rozich, J. D., Howard, R. J., Justeson, J. M., Macken, P. D., Lindsay,
M. F. & Resar, R. K. (2004) Standardization as a mechanism to
improve safety in health care: impact of sliding scale insulin protocol
and reconciliation of medications initiatives. Joint Commission
Journal on Quality and Safety, 30, 5–14.
29. DeCarolis, D. D., Leraas, M. C. & Rowley, C. (2005) Medication
reconciliation upon admit using an electronic medical record. Phar-
macotherapy, 25, 1505–1505.
30. Featherstone, P., Giles, P., Browne, M., et al. (2006) Improving
In-Patient Medicines Treatment Sheet Accuracy Within Hospitals. 11th
European Forum on Quality Improvement in Health Care, 26–28 April
2006, Prague, Czech Republic.
31. Dutton, K., Hedger, N., Wills, S., Brown, D. & Davies, P. (2003)
Prevent medication errors on admission. British Journal of Clinical
Governance, 8, 128–137.
32. Department of Health, Hospital Episode Statistics (2001–2002).
London: Department of Health.
33. Netten, A. & Curtis, L. (2007) Unit Costs of Health and Social Care.
Canterbury: PSSRU, University of Kent. http://www.pssru.ac.uk
(accessed 6 May 2008)
34. Nester, T. M. & Hale, L. S. (2002) Effectiveness of a pharmacist-
acquired medication history in promoting patient safety. American
Journal of Health-System Pharmacy, 59, 2221–2225.
35. Karnon, J., McIntosh, A., Bath, P., et al. (2006) A prospective hazard
and improvement analysis of medication errors in a UK secondary
care setting. Report for the NHS Patient Safety Programme.
36. Bates, D. W. (1997) The cost of adverse drug events in hospitalized
patients. Journal of the American Medical Association, 277, 307–311.
37. Pinilla, J., Murillo, C., Carrasco, G. & Humet, C. (2006) Case–control
analysis of the financial cost of medication errors in hospitalized
patients. European Journal of Health Economics, 7, 66–71.
38. Classen, D. C. (1997) Adverse drug events in hospitalized patients:
excess length of stay, extra costs, and attributable mortality. Journal of
the American Medical Association, 277, 301–306.
39. Schneider, P. J., Gift, M. G., Lee, Y. P., Rothermich, E. A. & Sill, B.
E. (1995) Cost of medication-related problems at a University hospi-
tal. American Journal of Health-System Pharmacy, 52 (21), 2415–
2418.
40. Douglas, J. & Larrabee, S. (2007) Bringing barcoding to the bedside.
Challenges and benefits of implementing a barcode point of care
medication administration system. Available at: http://www.
mihealthandsafety.org/presentations/larrabbee_douglas.ppt (last ac-
cessed 13 December 2007).
41. Hayward, R. A. & Hofer, T. P. (2001) Estimating hospital deaths due
to medical errors – preventability is in the eye of the reviewer. Journal
of the American Medical Association, 286, 415–420.
42. Bates, D. W., Teich, J. M., Lee, J., Seger, D., Kuperman, G. J.,
Ma’Luf, N., Boyle, D. & Leape, L. (1999) The impact of computerized
physician order entry on medication error prevention. Journal of the
American Medical Informatics Association, 6, 313–321.
43. Bates, D. W., Leape, L. L., Cullen, D. J., et al. (1998) Effect of
computerized physician order entry and a team intervention on pre-
vention of serious medication errors. Journal of the American Medical
Association, 280, 1311–1316.
Economic analysis of medicines reconciliation J. Karnon et al.
© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd
306