Full text of DOI:10.2165/00019053-200119080-00004

Pharmacoeconomics 2001; 19 (8): 819-829
1170-7690/01/0008-0819/$22.00/0
ORIGINAL RESEARCH ARTICLE
© Adis International Limited. All rights reserved.
The Cost Effectiveness of Stratified
Care in the Management of Migraine
Paul Williams,1 Andrew J. Dowson,2 Alan M. Rapoport3 and James Sawyer4
1
2
3
4
Genesis Pharma Strategies, Uxbridge, UK
Kings Headache Service, Kings College Hospital, London, UK
New England Center for Headache, Stamford, Connecticut, USA
Prism Ideas Ltd, Nantwich, UK
Objective: To examine the cost effectivess of a stratified-care regimen for pa-
tients with migraine – in which patients are stratified by severity of illness, and
then prescribed differing treatments according to level of severity – compared
with a conventional stepped-care approach.
Abstract
Design and methods: A decision analytic model was constructed to simulate a
controlled clinical trial in which patients with migraine receiving primary
medical care were randomly assigned to treatment under a stepped-care or a
stratified-care regimen. A health service payer perspective was adopted and the
time horizon was 1 year. Data inputs were: (i) the frequency and disability of
migraine, derived from population-based studies; (ii) disability level-specific
treatment response rates for over-the-counter analgesics,aspirin/metoclopram-
ide and zolmitriptan as the representative of high-end therapy obtained from an
international consensus opinion enquiry; and (iii) unit costs of healthcare ob-
tained from UK health service sources.
Main outcome measures and results: The estimated 1-year direct healthcare
costs per primary care patient with migraine were pound sterling (£) 156.82 for
stepped care and £151.57 for stratified care. Estimates of treatment response rates
were 40 and 71% for stepped and stratified care, respectively. The cost per suc-
cessfully treated attack was £23.43 for stepped care and £12.60 for stratified care.
Stratified care remained cost effective when tested in a wide range of one-way
sensitivity analyses, and probabilistic sensitivity analysis showed the cost effec-
tiveness of stratified care to be significant at the 3% level. Conditional confidence
analysis showed that the level of confidence in the cost effectiveness of stratified
care varied positively with the case mix, i.e. in populations where the proportion
of moderate and severely disabled patients with migraine was greater than 25%,
the cost effectiveness of stratified care remained statistically significant.
Conclusion: A stratified-care treatment strategy (including zolmitriptan as the
representative of high-end therapy) is a highly cost-effective method of managing
migraine in the primary care setting compared with stepped care, delivering im-
proved clinical outcomes at no additional cost.

820
Williams et al.
Migraine is a common neurological disorder
Optimal management of migraine requires a sys-
tematic evaluation of each patient and the develop-
ment of an individualised management plan. How-
ever, currently recommended treatment guidelines
for migraine typically involve a type of stepped
care, applied across or within attacks, where pa-
tients move through a series of medications from
nonspecific to specific therapies.[11] In stepped
care across attacks, patients initiate treatment with
1 medication for a series of attacks. If this nonspe-
cific treatment fails, physicians can step the patient
up to alternative, specific therapies for subsequent
attacks. In stepped care within attacks, patients ini-
tiate treatment with 1 medication for each attack.
If this treatment fails, they can step up their treat-
ment during their attack using escape medication.
Although stepped care may be an effective
strategy for patients with mild migraine, those with
greater treatment needs often find that initially
prescribed medications fail to relieve their symp-
toms and they may lapse from medical care. Strat-
ified care is an alternative strategy for managing
migraine. It is designed to match treatment to the
medical needs of each patient.[12,13] Patients can be
stratified according to their illness severity, as-
sessed as headache-related disability, for example
using the Migraine Disability Assessment (MIDAS)
Questionnaire,[14,15] a simple, widely available,[15]
5-item self-report questionnaire that can be com-
pleted easily and quickly by the patient. Treatment
is then prescribed to match the severity of illness.
In this way, patients who need specific therapies
receive them at their initial consultation, not after
a series of failed therapeutic efforts.
that affects 8 to 12% of the general population.[1,2]
Onset of migraine peaks in childhood and adoles-
cence[3] and prevalence peaks between the ages of
25 and 55 years, during the time of maximal work
and family commitments.[1,2] Prevalence is higher
in women than in men,[4] and in Caucasian than in
non-Caucasian races.[2]
Migraine exerts a significant burden on the in-
dividual in terms of pain, disability and reduced
quality of life.[5] This is compounded by the heter-
ogeneity of migraine. The frequency, duration,
symptomatology and resulting disability vary both
among and between individual patients, from attack
to attack and over time.[6]
Migraine also exerts a societal burden in terms
of direct and indirect costs of illness. Direct costs
comprise the cost of medical care. Indirect costs are
caused by absence from work, reduced productiv-
ity at work and disability in other roles. Most indi-
viduals with migraine have consulted a physician
at some time and take medications for the condi-
tion. Annual direct costs (adjusted to 1993 $US
values) range from $US12.5 million in Sweden to
$US1 billion in the US.[7,8] These figures proba-
bly underestimate present values, as they do not
include the cost of recently introduced migraine
therapies, such as the selective serotonin receptor
agonists (‘triptans’).
The indirect costs of migraine are substantial,
and are much higher than the direct costs. Work
losses related to reduced productivity are higher
than those related to work absence. Some of the
estimated annual costs (adjusted to 1993 $US val-
ues) are $US220 million in Australia, $US732
million in Canada, $US1.1 billion in Spain and
the UK, $US1.2 billion in The Netherlands and
$US13 billion in the US.[7,8] The results of these
studies show that the burden of migraine falls
disproportionately on patients and their employers,
while third-party payers shoulder less than 10% of
migraine-related economic costs. The overall cost
of migraine to society is comparable with that re-
ported for diabetes mellitus[9] and higher than that
reported for asthma.[10]
Stratified care for migraine (using the MIDAS
Questionnaire to grade patients’ illness severity)
was compared with stepped care across and within
attacks in a clinical trial, the Disability in Strategies
for Care (DISC) study.[16] The study showed
that stratified care, in which patients received
zolmitriptan 2.5mg or aspirin (acetylsalicylic acid)
and metoclopramide according to MIDAS grade,
was clinically and statistically significantly supe-
rior to the 2 stepped-care approaches, i.e. the across
and within attacks approaches. Significantly more
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

Cost Effectiveness of Stratified Care for Migraine
821
patients responded and became pain-free, and dis-
ability was significantly lower in the stratified
care group. The results showed that stratified
care optimised patients’ responses from the start
of treatment, whilst treatment benefits were delayed
in the stepped-care groups.
Stratified care, however, clearly means that
relatively more specific treatments will be pre-
scribed, as the more severely disabled patients re-
ceive these treatments earlier than with stepped care,
and specific treatments are more expensive. Can
the increased effectiveness of stratified care be ob-
tained without an additional burden on healthcare
costs? Is stratified care cost effective, compared
with stepped care? This is the focus of the study
reported in this paper.
The cost-effectiveness measure was the total
direct costs (as defined above, and including the
costs incurred by treatment failures) expected to be
incurred, divided by the expected number of effec-
tively treated migraine attacks, over the notional
year. Average cost-effectiveness ratios (for stepped
and stratified care, respectively) and the incremen-
tal cost-effectiveness ratio (for stratified care com-
pared with stepped care) were calculated as appro-
priate.
Structure of the Model
Three therapy levels were included in the model:
over-the-counter (OTC) analgesics, aspirin and meto-
clopramide, and high-end therapy as represented by
zolmitriptan. While an OTC arm was not included
in the DISC trial we felt it appropriate to include
this treatment here. Analgesics of this kindare widely
used, and indeed recommended as the starting point,
for example, by PRODIGY, the UK NHS comput-
erised decision support system for general practi-
tioners.[17]
Methods and Data
General Approach
The approach taken was to construct a decision
analytic model to simulate a controlled clinical trial
in which primary medical care patients with a di-
agnosis of migraine would be randomly assigned
to treatment under a stepped-care regimen across
attacks or a stratified-care regimen. A health ser-
vice payer perspective was adopted and the time
horizon was 1 year (no discounting was therefore
necessary).
The measure of effectiveness was the propor-
tion of migraine attacks expected to respond to treat-
ment during the notional year. The measure of cost
was the total of expected primary care consultation
costs, expected specialist referral and consultation
costs, and expected drug acquisition costs, over the
notional year. Inpatient costs were not included, as
the likelihood of admission for treatment of a mi-
graine attack is very low, especially in the context
of the UK National Health Service (NHS), which
was used to provide the cost base for this study
(any bias introduced by the exclusion of inpatient
costs can be presumed to act against stratified care,
given its greater effectiveness). The cost of per-
forming the MIDAS Questionnaire was considered
to be negligible. Indirect costs were not included.
Stepped Care
Figure 1 shows the structure of the stepped-care
regimen as represented in the model. Apatient con-
sults a primary care physician and migraine is di-
agnosed. OTC medication is advised, and the pa-
tient is told to return after treating a further 3 attacks.
At the second consultation, a judgement is made as
to whether the treatment has provided relief in 2 or
more of the 3 intervening attacks (the response cri-
terion used in the DISC trial). If so, the patient
continues on OTC medication and seeks care at a
less frequent rate (i.e. there is a greater number of
attacks between consultations; in the base case, it
was assumed, based on experience, that the con-
sultation rate would drop to 1 per 8 attacks). If not,
the patient ‘steps up’ to aspirin and metoclopram-
ide, and returns after treating a further 3 attacks.
At the third consultation, the physician again
makes a judgement as to whether the treatment has
provided relief on 2 or more of the 3 intervening
attacks. If so, the patient continues on aspirin and
metoclopramide and consults at a less frequent rate.
If not, the patient ‘steps up’ to high-end therapy
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

822
Williams et al.
(triptans) and returns after a further 3 attacks. The
same logic, using the same response criterion, is
followed at subsequent consultations, with pa-
tients whose migraines have not responded work-
ing their way through 3 different selective serotonin
receptors agonists and ultimately being referred for
specialist consultation (which, in the model, could
only occur at this time).
The 6 end-states, once reached, were assumed
to be absorbing (i.e. contain all patients) for the
remainder of the notional year, and patients were
assumed not to be lost to follow-up. For each end-
state, the model calculated the expected number of
attacks responsive to treatment, the expected num-
ber of primary care physician consultations, the ex-
pected number of attacks treated with a triptan, and
the expected likelihood of specialist referral/con-
sultation. These were multiplied by the calculated
end-state probabilities to provide the expected costs
and effectiveness for stepped care.
Stratified Care
Figure 2 shows the structure of the stratified-
care regimen as represented in the model. Apatient
consults a primary care physician, migraine is di-
agnosed, and a disability assessment is carried
out using the MIDAS Questionnaire. Patients in
MIDAS grade I (‘minimal or infrequent disability’)
follow the stepped-care approach, i.e. they may pro-
gress, according to clinical response, from OTC
analgesics through aspirin and metoclopramide
to a triptan and ultimately to specialist consult-
ation. Patients in MIDAS grade II (‘mild or infre-
quent disability’) enter this process at the aspirin and
metoclopramide stage, while patients in MIDAS
grades III (‘moderate disability’) and IV (‘severe
disability’) enter directly at the high-end therapy
stage. The criteria for when next to consult, and for
progressing through the treatments, were the same
in both the stratified- and stepped-care decision
trees.
There were 15 end-states (6 for MIDAS grade
I, 5 for MIDAS grade II, and 4 for MIDAS grades
III/IV). Once again, these end-states were assumed
to be absorbing and patients were assumed not to
be lost to follow-up. For each end-state, the model
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

Cost Effectiveness of Stratified Care for Migraine
823
calculated the expected costs and effectiveness for
stratified care as for stepped care.
Variables in the Model
Distribution of MIDAS Grades in the
Consulting Population (Case Mix)
Based on discussions with those who have clin-
ical experience with the MIDAS Questionnaire in
a UK general practice setting, it was assumed for
the base case that 5% of patients with migraine
seen in a primary medical care setting would fall
into MIDAS grade I, 25% into grade II and 70%
into grades III/IV. Because of the lack of actual
data on this crucial variable, we explored case mix
in some detail in the sensitivity analysis – see be-
low.
Number of Migraine Attacks Per Year,
by MIDAS Grade
Stewart et al.[18] carried out a validity study of the
MIDAS Questionnaire in which they used headache
diaries to assess the frequency of migraine attacks
overa 90-day period. TheMIDASGrade-specifichead-
ache frequencies shown in table I were extracted
from these data and provided to the present authors by
Dr. Kenneth Kolodner (personal communication).
These were scaled to provide figures applicable to
1 year (MIDAS grades III and IV were grouped
together, see table I).
Response Rates Specific to Each MIDAS Grade
for Each Level of Treatment
These were based on consensus expert opinion,
collected by means of a 2-round consensus opinion
(Delphi) enquiry designed to take account of Ev-
ans’[19] recommendations on the use of consen-
sus methods and expert panels in pharmacoecono-
mic studies. The participants were an international
group of 20 neurologists and headache specialists,
selected on the basis of their previous involvement
in the development and early use of the MIDAS
Questionnaire. Respondents were asked to make
judgements of clinical response based on their
knowledge and experience of providing treatment
at all 3 levels in the regimen, but were provided
with no specific data on which to base their judge-
ments. Estimates of attack response rate (table II)
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

824
Williams et al.
Table I. Number of headaches per year, by MIDAS grade (from
of the input variables on modelled cost, effective-
ness, and cost-effectiveness ratios. Then, a Monte
Carlo model was built to carry out a probabilistic
sensitivity analysis, i.e. a 17-way sensitivity anal-
ysis (the 16 input variables examined in the one-
way analyses, plus the effect of case mix) to assess
the impact on the cost effectiveness of stratified
care of changing all the input variables simulta-
neously. As part of this approach, conditional con-
fidence analysis[22] was used to explore in more
detail the impact of the level of disability (the case
mix) on the cost effectiveness of stratified care in
the consulting population.
analyses carried out on data collected by Stewart et al.[18]
)
MIDAS grade
Number of migraine Calculated annual
attacks in a 90-day number of migraine
period (mean SD) attacks used in the
model (mean SD)
I
II
III
IV
1.85 2.52
1.98 2.62
3.01 2.65
6.66 5.32
5.11 4.36
7.49 10.22
8.02 10.61
12.22 10.74
27.01 21.59
20.74 17.70
III/IV combined
MIDAS = Migraine Disability Assessment Questionnaire; SD =
standard deviation.
were used to calculate the likelihood of patients
responding in at least 2 of 3 attacks, for use in the
base case model. If the probability of responding
in 1 migraine attack is p, then the probability of
responding in at least 2 out of 3 attacks – assuming
independence between attacks – is p3 + 3(p2[1–p])
= 3p2 – 2p3. (MIDAS grades III and IV were again
grouped together). The interquartile ranges (table
II) were used to represent the extent of uncertainty
in the sensitivity analyses (as suggested by Rowe
et al.[20]).
The input variables were modelled as follows:
Per-Grade Mean Number of Migraine
Attacks Per Year
The means and standard deviations calculated
by Dr Kenneth Kolodner (personal communica-
tion) on the basis of the Stewart et al.[18] data trans-
formed to be appropriate for 1-year rather than 90-
day data (table I) were used as parameters to define
γ distributions (the γ distribution was selected be-
cause of its appropriateness for right-skewed data
Unit Costs of Health Service Consumption
The UK NHS costs were used. For the base case,
the cost of a primary care consultation and special-
ist referral/consultation were set at pounds sterling
(£) 16 and £93, respectively.[21] The cost of treating
a migraine attack with a selective serotonin recep-
tor agonist was £6 for the base case (the cost of 1.5
tablets of zolmitriptan 2.5mg); OTC analgesics and
aspirin and metoclopramide were considered to be
of negligible cost and therefore were not included
in the analysis.
Table II. Estimates of response rates specific to each MIDAS grade
for each level of treatment: medians and interquartile ranges from
the second (final) round of the consensus opinion enquiry
Midas
grade
Treatment
Response rates (%)
median interquartile
range
I
OTC analgesics
Aspirin/metoclopramide
Zolmitriptan
OTC analgesics
Aspirin/metoclopramide
Zolmitriptan
OTC analgesics
Aspirin/metoclopramide
Zolmitriptan
OTC analgesics
Aspirin/metoclopramide
Zolmitriptan
47
57
69
34
46
69
21
34
66
15
24
61
18
29
64
33–57
46–65
62–76
24–47
37–45
62–75
15–28
27–47
59–73
7–21
18–32
53–69
11–25
21–41
56–71
II
Statistical Analysis
III
IV
Estimates for cost indices, effectiveness indices
and cost-effectiveness ratios were first estimated
for the base case, following which extensive sensi-
tivity analysis was undertaken.
III/IV
combined
OTC analgesics
Aspirin/metoclopramide
Zolmitriptan
Sensitivity Analysis
A set of 16 one-way sensitivity analyses was
carried out to explore the impact of changes in each
OTC = over-the-counter; MIDAS = Migraine Disability
Assessment Questionnaire.
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

Cost Effectiveness of Stratified Care for Migraine
825
Table III. Expected units and costs of health service consumption, per patient per year, as estimated by the base-case model
Parameter
Stepped care
Stratified care
Difference (stratified
care – stepped care)
Number of consultations with primary care physician
Cost of consultations with primary care physician
Number of specialist referrals/consultations
Cost of specialist referrals/consultations
Number of attacks treated with selective serotonin
receptor agonists
4.90
£78.44
0.23
£21.32
9.51
3.65
£58.34
0.02
£1.68
15.26
–1.25
–£20.10
–0.21
–£19.64
5.75
Cost of selective serotonin receptor agonists
Total health service costs
£57.06
£156.82
£91.55
£151.57
+£34.49
–£5.24
£ = pounds sterling.
without negative values). The distributions were
sampled randomly in each iteration of the model
using the algorithm given by Morgan and Henr-
ion.[23]
Results
The Base Case
Table III shows the expected units and costs of
health service consumption, per patient per year, as
estimated by the model. In the base case – i.e. with
input values as given in the previous section – the
stratified-care regimen resulted in fewer primary
care consultations and fewer specialist consultations
than the stepped-care regimen, resulting in an ex-
pected saving to the health service (in 1 year per
patient) of £20.10 and £19.64, respectively. These
savings were partially offset by the greater number
of attacks treated with a selective serotonin recep-
tor agonist under stratified care, which incurred a
correspondingly greater drug treatment cost. Even
so, with base case input values the total health ser-
vice cost was £5.24 per patient per year lower un-
der a stratified-care regimen than under a stepped-
care regimen (table III). These estimates for cost
saving are likely to be underestimates, as it might
be expected that the enhanced efficacy due to strat-
ified care would result in reduced healthcare ex-
penditure on comorbid conditions.
Table IV shows the mean number of migraine
attacks per year (the mean of the values in table I
weighted by the frequencies of the MIDAS grades
assumed for the base case model, as described ear-
lier). Under stepped care, the model estimated 6.69
attacks to be responsive to treatment (a 40% attack
response rate), whereas under stratified care, the
estimated number of treatment-responsive attacks
was 12.02, or 71% of attacks over the notional year.
Per-Grade Treatment Response Rates
In each iteration of the model, values were
sampled randomly from a uniform distribution
between the twenty fifth and seventy fifth percen-
tiles obtained from the consensus opinion enquiry
study (table II). This approach was adopted rather
than an algorithm making use of the central ten-
dency – e.g. sampling from a triangular distribu-
tion – as this provided a more robust test of the cost
effectiveness of stratified care.
Unit Costs of Health Service Consumption
In each iteration of the model, values were sam-
pled randomly from a uniform distribution between
the extremes of 50% of the base case value.
Post-Recovery Consultation Rate
In each iteration of the model, values were sam-
pled randomly from a uniform distribution between
the extremes of the preconsultation rate (1 consult-
ation per 3 attacks) and 1 consultation per year (1
consultation per 16.89 attacks).
Case mix
In each iteration of the model, the proportions
of patients in each MIDAS grade were sampled
randomly from uniform distributions between 0 and
1, subject to the constraint that the total of the 3
proportions (for Grades I, II, III/IV) should sum to 1.
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

826
Williams et al.
For stepped care, the model estimated that a health
Carlo model of 17-way analysis, as outlined in
the Methods section. Following Critchfield and
Willard,[22] these exact probabilities can be interpre-
ted as levels of confidence in the result. Therefore,
within the confines of the model structure and the
ranges of input values tested in the analysis, we can
be:
service expenditure of £23.43 was required to de-
liver 1 successfully treated migraine attack [£156.82
(table III) divided by 6.69 (table IV)]. For stratified
care, however, an expenditure of only £12.60 was
required to deliver the same outcome [£151.57 (ta-
ble III) divided by 12.02 (table IV)].
The reciprocal of the cost-effectiveness ratios
represents the number of successful outcomes that
can be ‘purchased’ for a given health service ex-
penditure. Table IV shows that, for stepped care,
4.27 successfully treated migraine attacks were ‘pur-
chased’ for £100 of health service expenditure [6.69
(table IV) divided by £156.82 (table III) × 100]. For
stratified care by comparison, the same expendi-
ture purchased 7.93 successful outcomes [12.02
(table IV) divided by £151.57 (table III) × 100], an
increase of 80%.
• 33.6% confident that stratified care will be cost
saving compared with stepped care
• 99.5% confident that stratified care will be more
effective than stepped care
• 96.4% confident that stratified care will be cost
effective, compared with stepped care: that is,
the cost effectiveness of stratified care can be
considered significant at the p = 0.036 level.
In the Monte Carlo analysis described in the
Methods section, the case mix was allowed to vary
widely (the probability of each of the grades was
allowed to range between 0 and 1, subject only to
the constraint that they should sum to 1). Even though
the impact of case mix on the cost effectiveness of
stratified care has been examined in this way, it is
appropriate to examine this in more detail to dis-
cover the relevance of the stratified care approach
in differing clinical populations (especially as the
case mix in the model base case was based on as-
sumption rather than data).
To do this, we used conditional confidence anal-
ysis, a technique in which the variable of interest
is held constant, while the values of all the other
input variables are allowed to vary in the Monte
Carlo model. It is thus analogous to the ‘step down’
approach to examining the impact of a predictor
variable in a multiple regression analysis. We used
Calculating an incremental cost-effectiveness
ratio for stratified care was unnecessary as strati-
fied care was both less costly and more effective
than stepped care (with base case input values).
Sensitivity Analyses
16 one-way sensitivity analyses were carried out.
Stratified care remained dominant over stepped care
in 10, while in the other 6, stratified care was cost
effective, i.e. the incremental cost-effectiveness
ratio was lower than the cost-effectiveness ratio for
stepped care. Details of these one-way sensitivity anal-
yses are available on request from the first author.
Table V shows the exact probabilities result-
ing from a run of 10 000 iterations of the Monte
Table IV. Effectiveness of, and cost-effectiveness ratios for, stepped and stratified care, as estimated by the base-case model
Parameter
Stepped care
Stratified care
Difference (stratified care –
stepped care)
Number of migraine attacks per yeara
Number of migraine attacks per year responding to treatment 6.69
Attack response rate (%)
16.89
16.89
12.02
71
+5.33
31
40
Health service costs per successfully treated attack
Number of successfully treated attacks that can be
‘purchased’ with £100 of health service expenditure
£23.43
4.27
£12.60
7.93
Stratified care dominant
Stratified care dominant
a
The figures in this row are an input to, rather than an output from, the model.
£ = pounds sterling.
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

Cost Effectiveness of Stratified Care for Migraine
827
Table V. Exact probabilities resulting from 10 000 iterations of the Monte Carlo model
Cost
Probability that health service costs are less with stratified care than with
stepped care
Probability that effectiveness is greater with stratified care than with
stepped care
0.336
0.995
Effectiveness
Cost-effectiveness
ratio
Probability that stratified care is dominant over stepped care
0.335
0.629
Stratified care
cost effective
0.964
Probability that stratified care, while not dominant, is cost effective
compared with stepped carea
Probability that stepped care, while not dominant, is cost effective
compared with stratified careb
Probability that stepped care is dominant over stratified care
0.036
0
Stratified care
not cost
effective 0.036
a
b
The incremental cost-effectiveness ratio for stratified care is less than that for stepped care.
The cost-effectiveness ratio for stepped care is less than the incremental cost-effectiveness ratio for stratified care.
the probability of MIDAS grades III/IV as our in-
dex of case mix (i.e. the expected level of disability
increases as this probability increases), and we car-
ried out 9 runs of the model (1 for each probability
decile) of 5000 iterations each.
Figure 3 shows that, as might be expected, the
level of confidence in the cost effectiveness of strat-
ified care (as compared with stepped care) increased
as the expected level of disability increased. Even
at the lowest disability level, however, the confi-
dence in the cost effectiveness of stratified care
was greater than 90%. If the usual 95/5% level
is taken as the threshold of acceptability/signifi-
cance, then stratified care can be considered sig-
nificantly cost effective in populations where the
probability of MIDAS grades III/IV is greater than
25%.
the data. As in all modelling studies, we attempted
to keep these limitations to a minimum.
The imposition of a clinical trial structure does
itself impose limitations on the interpretation of
the results. For instance, in clinical practice, some
patients in MIDAS grade I may already have tried
OTC analgesics (and would thus step up immedi-
ately), while some patients in MIDAS grade IV
may require particular care programmes tailored to
their high disability level.
The data used in the model were, where possi-
ble, derived from published sources. There were 3
principal data limitations. First, there are no avail-
able data on attack response rates for each of the 3
treatment levels (low, intermediate and high-end)
specific for MIDAS grades, so these were based
on expert opinion collected by means of a 2-round
consensus opinion enquiry (in which high-end treat-
ment was represented by zolmitriptan). In accord-
ance with published recommendations for such
studies,[18] the participants were acknowledged ex-
perts in their field, and steps were taken to avoid
attrition between the rounds (in fact, there was no
attrition at all). The material provided to the ex-
perts was standardised and contained no anchoring
information, and the feedback comprised both in-
dividual responses (anonymised except for the re-
cipients’ own) and group responses. Although
there are no data with which the results of the
consensus opinion enquiry can be compared di-
rectly (if there were, the enquiry would not have
been necessary), we would expect a reasonable
Discussion
Within the limitations of the study, stratified
care was shown to be a cost-effective approach to
the management of migraine in the primary care
setting, compared with the usual stepped-care ap-
proach. Indeed, in the base case, stratified care was
cost saving (albeit of a relatively small sum) with
resources spent producing some 80% more benefit
than the same resource spent in a stepped-care set-
ting.
All modelling studies have limitations – those
imposed by the structure of the model and the in-
herent assumptions, as well as those imposed by
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

828
Williams et al.
1.00
0.99
0.98
0.97
0.96
0.95
0.94
0.93
0.92
0.91
0.90
be effective are more likely to remain in treatment.
This being so, we believe it reasonable to assume
that the insensitivity of the principal result to vari-
ation in treatment response rates (both in the one-
way and 17-way sensitivity analyses) is a proxy for
robustness of the result to reduced compliance with
follow-up.
Given these limitations, the results of the sensi-
tivity analyses are especially important. The cost
effectiveness of stratified care was robust when ex-
amined in a set of 16 one-way sensitivity analyses
(not including the relative frequency of MIDAS
grades). Stratified care remained dominant in 10 of
these, and was cost effective in the remaining 6
analyses, with the incremental cost-effectiveness
ratio well below the average cost-effectiveness ra-
tio for stepped care. In the probabilistic sensitivity
analysis, all input variables (including the relative
frequency of MIDAS grades) were allowed to vary
simultaneously. Conservative assumptions were made
in the design of the Monte Carlo model – for exam-
ple, uniform distributions were used to model the
treatment response rates, and the case mix was al-
lowed to vary without constraint (except that the
frequencies of the MIDAS categories were required
to sum to 1). This latter assumption meant that, in
this analysis, the expected proportion of MIDAS
grades III/IV was 33% (each of the 3 grades had an
equal probability of selection). On this basis, there-
fore, the cost effectiveness of stratified care was
statistically significant even when assessed on a
population considerably less severe than that ex-
pected in a general practice setting – i.e. the signif-
icance level p = 0.036 is a conservative estimate.
This was confirmed and extended in the conditional
confidence analysis, where the level of confidence
(in the economic superiority of stratified care) was
shown to vary positively with case mix, and to con-
form with the conventionally accepted level of sig-
nificance in populations where the proportion of
MIDAS grades III/IV was greater than 25%.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Case mix: probability of MIDAS grades III/IV
Fig. 3. The relationship between case mix (as measured by the
probability of MIDAS grades III/IV) and level of confidence in the
cost effectiveness of stratified care. MIDAS = Migraine Disability
Assessment Questionnaire.
correspondence between the experts’ estimates for
MIDAS grades III/IV and results from migraine clin-
ical trials, given that many migraine clinical trials
are conducted at headache speciality centres and
tend to enrol patients with more severe migraine.
In the 5 clinical trials of zolmitriptan 2.5 to 5mg
reported in the product labelling information, the re-
sponse rate (defined in the same way as in the con-
sensus opinion enquiry) ranged from 59 to 67%.[24]
Table II shows that the interquartile range for the
relevant rates from consensus opinion was 56 to
71%, a very satisfactory level of correspondence
which provides a degree of convergent validity.
The second principal data limitation is the ab-
sence of survey information on the relative fre-
quency of the MIDAS grades in patients with
migraine receiving primary care. We made an as-
sumption based on clinical experience of treating mi-
graine in general practice (70% of patients seeking
treatment would fall into MIDAS grades III/IV),
but even so, we paid special attention to this case
mix variable in the sensitivity analysis.
The third data limitation is the assumption that
no patients were lost to follow-up. Once again, as
far as we are aware, there are no data that address
this point. It would be reasonable to assume, how-
ever, that effectiveness and compliance with follow-
up are related – i.e. patients who find their treatment to
The results of any modelling study are im-
mensely strengthened if they can be confirmed in
the real world. While the DISC study[16] showed
the clinical superiority of stratified care compared
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)

Cost Effectiveness of Stratified Care for Migraine
829
10. Smith DH, Malone DC, Lawson KA, et al. A national estimate
of the economic costs of asthma. Am J Respir Crit Care Med
1997; 156: 787-93
with stepped care, it was not suitable for examining
direct costs as these were largely protocol driven.
11. MIPCA. Migraine management guidelines: a strategy for
the modern management of migraine. In: Eisenstadt, editor.
Guidelines: summarising clinical guidelines for primary care.
Berkhamsted: Medendium 1998; 4: 147
12. Lipton RB, Stewart WF. Clinical applications of zolmitriptan
(Zomig™, 311C90). Cephalalgia 1997; 17 Suppl. 18: 53–9
13. Lipton RB. Disability assessment as a basis for stratified care.
Cephalalgia 1998; 18 Suppl. 22: 40–6
14. Stewart WF, Lipton RB, Whyte J, et al. An international study
to assess the reliability of the Migraine Disability Assessment
(MIDAS) score. Neurology 1999; 53: 988–94
15. MIDAS questionnaire. Accessed from: URL: http: //www.midas-
migraine.net/us/question/default.asp?p=question [Accessed
2001 May 31]
16. Lipton RB, Stewart W, Stone A, et al. Stratified care is more
effective than step care strategies for migraine: results of the
Disability In Strategies for Care (DISC) Study. JAMA 2000;
284: 2599-605
17. Migraine. Accessed from: URL: http://www.prodigy.nhs.uk/
guidance/crs/migraine.htm [Accessed 2001 May 31]
18. Stewart WF, Lipton RB, Kolodner KB, et al. Validity of the
Migraine Disability Assessment (MIDAS) score in compari-
son to a diary-based measure in a population sample of mi-
graine sufferers. Pain 2000; 88: 41-52
Conclusions
Using a decision analytic model designed to
simulate a controlled clinical trial, stratified care
(which included zolmitriptan as the representative
of high-end therapy) was shown to be a highly cost-
effective method of managing migraine in the pri-
mary care setting compared with stepped care, de-
livering improved clinical outcomes at no additional
cost.
Acknowledgements
This study was supported by a grant from AstraZeneca
Pharmaceuticals. PW, AJD and AMR have undertaken con-
sultancy work on behalf of a number of pharmaceutical
companies. At the time of the study, JS was an employee of
AstraZeneca.
19. Evans C. The use of consensus methods and expert panels in
pharmacoeconomic studies: practical applications and meth-
odologicalshortcomings. Pharmacoeconomics1997;12:121–9
20. Rowe G, Wright G, Bolger F. Delphi: a re-evaluation of
research and theory. Technological Forecasting and Social
Change 1991; 39: 235-51
21. Netten A, Dennett J, Knight J. Unit costs of health and social
care. Canterbury: Personal and Social Services Research Unit,
University of Kent at Canterbury, 1998
22. Critchfield GC, Willard KE. Probabilistic analysis of decision
trees using Monte Carlo simulation. Med Decis Making 1986;
6: 85-92
23. Morgan MG, Henrion M. Uncertainty: a guide to dealing with
uncertainty in quantitative risk and policy analysis. Cambridge:
Cambridge University Press, 1990
References
1. Stewart WF, Simon D, Schechter A, et al. Population variance
in migraine prevalence: a meta-analysis. J Clin Epidemiol
1995; 48: 269–80
2. Scher AI, Stewart WF, Lipton RB. Migraine and headache: a
meta-analytic approach. In: Crombie IK, editor. Epidemiol-
ogy of pain. Seattle: IASP Press, 1999; 159–70
3. Stewart WF, Linet MS, Celentano DD, et al. Age- and sex-specific
incidence rates of migraine with and without visual aura. Am
J Epidemiol 1991; 134: 1111–20
4. Johannes CB, Linet MS, Stewart WF, et al. Relationship of
headache to phase of the menstrual cycle among young
women: a daily diary study. Neurology 1995; 45: 1076–82
5. Dahlöf CGH, Solomon GD. The burden of migraine to the in-
dividual sufferer: a review. Eur J Neurol 1998; 5: 525–33
6. Stewart WF, Shechter A, Lipton RB. Migraine heterogeneity:
disability, pain intensity, and attack frequency and duration.
Neurology 1994; 44 Suppl. 4: 24–39
24. ZOMIG Professional Information Brochure. Accessed from: URL:
http://www.astrazeneca-us.com/pi/ZM1108.pdf [Accessed 2001
May 31]
7. Ferrari MD. The economic burden of migraine to society. Phar-
macoeconomics 1998; 13: 667–76
8. Hu XH, Markson LE, Lipton RB, et al. Burden of migraine in
the United States: disability and economic costs. Arch Intern
Med 1999; 159: 813–8
9. Thom TJ. Economic costs of neoplasms, arteriosclerosis, and
diabetes in the United States. In Vivo 1996; 10: 255-9
Correspondence and offprints: Dr Paul Williams, Genesis
Pharma Strategies, PAREXELMMS, River Court, 50 Oxford
Road, Denham, Uxbridge, Middlesex UB9 4DN, UK.
© Adis International Limited. All rights reserved.
Pharmacoeconomics 2001; 19 (8)