Cost-Effectiveness Analysis of Stratified Versus Stepped Care Strategies for Acute Treatment of Migraine

Cost-Effectiveness Analysis ofStratified Versus Stepped CareStrategies for Acute Treatmentof MigraineThe Disability in Strategies for Care (DISC) Study

Mark Sculpher,1 David Millson,2 David Meddis3 and Lynne Poole3

1 Centre for Health Economics, University of York, York, United Kingdom2 Department of Medicines Management, Keele University, Keele, Staffordshire, United Kingdom3 AstraZeneca, Macclesfield, United Kingdom

Abstract Background: The Disability in Strategies for Care (DISC) study was the firstlarge randomised controlled trial to compare alternative treatment strategies inthe acute treatment of migraine. With 835 patients in its intention-to-treat efficacyanalysis, DISC compared a stratified care strategy, where initial therapy wasbased on clinical need as determined by the Migraine Disability Assessment Scale(MIDAS) and two stepped care strategies (across attacks and within attacks),where first-line therapy with a simple combination analgesic was escalated, ifresponse had been inadequate, to zolmitriptan, a migraine-specific therapy.Objective: To report on the cost effectiveness of these three strategies from asocietal perspective.Study design and methods: A cost-effectiveness analysis was undertaken usingdata from the DISC study, and including both health service and productivitycosts. Data were collected prospectively on drug usage (main therapy and rescuemedication); resource use associated with adverse events was estimated by aclinician blinded to treatment strategy. Health service resource use was costedusing UK unit costs (1999 to 2000 values). Data were collected using diary cardson the amount of time patients lost from work, and on reduced effectiveness atwork, due to a migraine attack. This facilitated an estimate of the productivitycosts associated with the treatment strategies. To assess cost effectiveness, thedifferences in costs between the strategies were related to the two primary out-come measures in the trial: headache response 2 hours after initial therapy anddisability-adjusted time during the first 4 hours after initial therapy.Results: Although the mean health service cost was higher in the stratified caregroup (mean over 6 attacks of £28.25 versus £11.74 and £23.15 in the steppedcare across attacks group and within attacks group, respectively), mean produc-tivity costs over 6 attacks were lower in the stratified group (£112.22 versus£144.70 and £127.53). The total mean cost over six attacks was, therefore, lowestin the stratified care group (£138.95 compared with £157.19 in the stepped care

ORIGINAL RESEARCH ARTICLE Pharmacoeconomics 2002; 20 (2): 91-1001170-7690/02/0002-0091/$25.00/0

© Adis International Limited. All rights reserved.

across attacks group and £148.53 in the stepped care within attacks group), al-though these differences did not reach statistical significance. In terms of head-ache response, stratified care was statistically significantly more effective thanboth forms of stepped care. Using disability-adjusted time, stratified care wasstatistically significantly more effective than stepped care across attacks, but notagainst stepped care within attacks.Conclusion: Given its lower mean costs and higher mean effectiveness, a strat-ified care strategy, which included zolmitriptan, was the dominant strategy andwas unequivocally more cost effective from a societal perspective than eitherstepped care strategy. When the uncertainty around these means was considered,stratified care had the highest probability of being cost effective.

A range of treatment options is now available tomanage migraine.[1-3] Given that alternative thera-pies vary in their efficacy, adverse effects andcost, clinical guidelines have suggested severalstrategies for selecting and sequencing acute ther-apies.[2,3] Three strategies in particular haveemerged in these guidelines: stepped care acrossattacks, stepped care within attacks and stratifiedcare.[2-5]

With stepped care across attacks, all patientsbegin with a simple or combination analgesic andif, after treating a number of attacks, the patient isunhappy with their treatment they contact their cli-nician to change therapy. This process continuesuntil an acceptable therapy is identified. Steppedcare within attacks involves the patient starting offeach attack with a simple or combination analgesicbut, if satisfactory results are not achieved by aspecific time point during the attack (usually 2hours after onset), the patient takes another medi-cation, usually migraine-specific. With stratifiedcare, the choice of initial medication follows anassessment of each individual patient’s needs, withthose patients experiencing the greatest disabilityduring attacks usually receiving migraine-specifictherapy.[3-5]

The Disability in Strategies for Care (DISC)study was a unique trial set up to compare the twostepped care strategies (i.e. stepped care across at-tacks and stepped care within attacks) with strati-fied care.[6] This randomised controlled trial foundthat a stratified care treatment strategy, which in-cluded zolmitriptan as the migraine-specific ther-

apy, was more effective, in terms of headache re-sponse and disability time, than either stepped careapproach.[6] A modelling study has recently beenpublished simulating the economic results of theDISC study.[7] This paper uses patient-level datacollected in the DISC trial, and reports on a cost-effectiveness analysis of the alternative strategiesin which their differential health service and pro-ductivity costs are related to differences in out-comes.

Methods

Clinical Trial Design

Full details of the design of the DISC trial havebeen published elsewhere.[6] In brief, the study wasa randomised, parallel group, open label, multiplemigraine attack trial undertaken in 88 centres in 13countries. Patients treated up to six of their attacksusing the treatment strategy to which they wererandomised. If a patient treated fewer than six at-tacks, they remained in the trial for 6 months.

At baseline, all patients completed the MigraineDisability Assessment Scale (MIDAS) question-naire, a validated instrument measuring migrainerelated disability.[8-10] The instrument was used tograde each patient entering the trial: Grade I (littleor infrequent disability), Grade II (mild or infre-quent disability), Grade III (moderate disability)and Grade IV (severe disability). Only patientsgraded II to IV were included in the study. A rangeof other inclusion criteria included being aged 18to 65 years and fulfilling the International Head-

92 Sculpher et al.

© Adis International Limited. All rights reserved. Pharmacoeconomics 2002; 20 (2)

ache Society (IHS) criteria for migraine, with orwithout aura.[6] Exclusion criteria included a med-ical or psychiatric condition increasing risk whenexposed to trial medications or interfering with ef-ficacy and safety assessments; a history of basilar,ophthalmoplegic or hemiplegic migraine; and ahistory of heart disease, stroke or significant med-ical illness.

Eligible patients were randomised to 1 of 3groups:

Stratified care. Grade II patients received aspi-rin [acetylsalicylic acid] (800 to 1000mg) plusmetoclopramide 10mg and patients graded III orIV received zolmitriptan 2.5mg to treat all mi-graine attacks.

Stepped care across attacks. Patients treatedtheir first three attacks with aspirin (800 to1000mg) plus metoclopramide 10mg. Patientswith an unsatisfactory headache response duringthe first three attacks (reduction in headache sever-ity from severe or moderate to mild or no headacheat 2 hours in fewer than two attacks, i.e. one or zeroattacks) were instructed to step up to zolmitriptan2.5mg for the following three attacks.

Stepped care within attacks. Patients startedtreatment with aspirin (800 to 1000mg) plus meto-clopramide 10mg for all attacks. For each attack,step up to zolmitriptan 2.5mg was permitted if aheadache response (reduction in headache severityfrom severe or moderate to mild or no headache)was not achieved after 2 hours.

Patients in all three groups were asked to treatonly moderate or severe headaches. Patients wererequested to avoid rescue medication (drugs usedif study medication failed to improve symptomssufficiently) in the first 4 hours. However, if theyexperienced persistent or recurrent migraine be-tween 4 and 24 hours after their initial medicationthey were permitted to take an additional dose ofstudy medication or another rescue medication, al-though the latter could not be another triptan. Pa-tients kept diary cards to detail each migraine at-tack, and these were reviewed after the third andsixth attack. A total of 835 patients were random-ised and included in the intention-to-treat analysis:

279, 271 and 285 in the stratified, stepped careacross attacks and stepped care within attackgroups, respectively.

Effectiveness

Two primary clinical endpoints were includedin the trial. These have been used as alternativemeasures of effectiveness in this cost-effectivenessanalysis. The first was headache response rate at 2hours over six attacks, where a ‘response’ was de-fined as a reduction in pain intensity from severeor moderate at the time of drug administration tomild or absent. The second was disability-adjustedtime over 4 hours. This was based on each patientrecording their level of functioning pre-treatmentand at 1, 2 and 4 hours post-treatment on a visualanalogue scale running from 0 (complete lack offunction) to 100 (complete function). The area un-der the curve (AUC) from 0 to 4 hours was calcu-lated for each patient where maximum functionwould be 400 and no function would be 0. This wastransformed to disability-adjusted time (h) asAUC/100 for each patient, where the maximumdisability-adjusted time over each attack was 4hours (i.e. 4 hours without disability) and whereanything less than this was a reflection of a pa-tient’s disability. Results are presented as disabil-ity-adjusted time per six attacks [i.e. a maximumof 24 hours (4 hours × six attacks) without dis-ability].

Costs

A societal perspective was adopted for the cost-effectiveness analysis. All important resourcesconsumed in the DISC trial were costed using 1999to 2000 UK unit costs. The following cost catego-ries were considered:

Study drugs. All drug acquisition costs werebased on undiscounted prices in the March 2000British National Formulary.[11] Given their verylow acquisition price, the acquisition cost of aspi-rin and metoclopramide has not been estimated. Inthe case of stratified care and stepped care betweenattacks, the initial dose of zolmitriptan was notrecorded by the patient so it has been assumed that

Acute Treatment of Migraine 93


a standard 2.5mg dose was taken as required by theprotocol. All other doses of zolmitriptan were re-corded.

Rescue medication. Based on data recorded bypatients in their diary cards, a cost was attached toeach rescue medication. This was possible in themajority (88%) of cases. For those where costingwas not possible (e.g. because the drug is not mar-keted in the UK), a cost was imputed based on theaverage of those that were costed.

Adverse events. Resource implications of ad-verse events were not recorded in the trial, althoughdetails of the events themselves were. Therefore,one of the authors (D. Millson), blinded to the treat-ment strategy to which a patient was randomised,considered all serious or severe adverse events,or those leading to withdrawals from the trial. Hedescribed the most likely form of pharmaceuticaland non-pharmaceutical management for eachevent and these were costed using published unitcosts.[12-14]

Productivity costs. In a cost-effectiveness anal-ysis conducted from a societal perspective, it is ap-propriate to consider the costs that are incurred asa result of patients forgoing their usual activitiesdue to a migraine attack.[15] Two types of activityare affected: paid work and other activities (e.g.unpaid work such as housework, and leisure). Asis the case in most economic evaluations from asocietal perspective,[16] the value of activities otherthan paid work is assumed to be reflected in themeasure of effectiveness. Therefore, only the costsof foregone paid working time were included in theanalysis. These productivity costs may be incurredby the patient or by their employer, but they repre-sent a cost to society regardless of who incurs it.

Productivity costs can result from patients hav-ing to take time off work as a result of their mi-graine or from reduced effectiveness while at work.Both of these effects were captured through the useof the lost work equivalent (LWE), which is de-fined as:

LWE = LWT + (working hours x [100 – %effectiveness])

where LWT is lost working time, working hours isthe number of hours at work during which a patientwas affected by a migraine and effectiveness is ameasure of effectiveness at work measured on a4-point scale from 0 (fully effective) to 1 (not at alleffective). The second and third points on this scalehave been used to represent 75 and 25% effective-ness at work, respectively. Patients provided theinformation necessary to calculate LWEs for eachattack in their diary cards. To translate theLWE into monetary terms, each LWE hour wasvalued using the average UK wage rate.[17] It hasbeen assumed that the periods of time affected bya migraine attack are too short to acquire pre-viously unemployed replacement workers (i.e. theabsence due to sickness is shorter than the frictionperiod).[18]

Cost-Effectiveness Analysis

For each patient in the trial total cost and effec-tiveness over six attacks was calculated. For thosepatients who treated fewer than six attacks, costshave been scaled up on a pro rata basis so that theyare presented as being over the equivalent of sixattacks. For each of the three randomised groups,mean cost and effectiveness per patient was calcu-lated. 95% confidence intervals are presented forthese means which, given the skewed nature of thedata, are based on non-parametric bootstrappingusing the bias-corrected approach.[19] Given theshort time horizon in the study (6 months or sixattacks, whichever was the shorter), discountingwas not required.

The cost-effectiveness analysis was undertakenin two stages. The first is deterministic and relatesthe mean costs and effectiveness of the alternativestrategies using standard decision rules.[20] Thesecond stage is a stochastic analysis which reflectsthe uncertainty around the mean cost and effective-ness estimates. This extended analysis estimatesthe probability of each of the three managementstrategies being more cost effective than the othertwo, requiring the following to be observed: (i) thestrategy has a lower cost and is more effective thanboth alternative strategies; (ii) the strategy is more

94 Sculpher et al.


costly and more effective than one or both of theothers, but the extra cost of achieving the addi-tional effectiveness (the incremental cost-effec-tiveness ratio) is considered acceptable by decisionmakers; (iii) one or both of the other strategies ismore costly and more effective but its incrementalcost-effectiveness ratio is considered too high bydecision makers. The analysis is presented in theform of a cost-effectiveness acceptability curvewhich shows the probability of each of the threestrategies being more cost effective than the othertwo for a range of alternative maximum values ofthe incremental cost-effectiveness ratio that deci-sion makers are willing to pay.[21,22] For a societalcost-effectiveness analysis, these willingness topay values should reflect the public’s valuation ofa unit of health gain. Cost-effectiveness accept-ability curves are being increasingly used in ap-plied economic evaluations[23,24] and advocated asa method for presenting uncertainty in the resultsof studies.[25]

Results

Resource Use

Table I presents summary statistics for keyresource parameters collected in the trial. Notsurprisingly, the amount of zolmitriptan used (in-cluding as a rescue medication) was higher in thegroup receiving stratified care than in the other twogroups: mean of 15.42mg per patient over six at-tacks compared with 3.65mg and 10.88mg in thestepped across attacks and stepped within attacksgroups, respectively. The other variable shown inthe table is the number of lost LWEs for the threegroups over the six attacks. LWEs were lowest inthe stratified care group (mean 11.19), followed by

the stepped care within attacks (mean 12.72), withstepped care across attacks associated with thehighest LWEs (mean 14.43).

The other key resource use in the study relatedto the use of rescue medication and the incidenceof adverse events. As regards rescue medication(including zolmitriptan), 39% of attacks in patientsrandomised to stratified care required rescue med-ication, compared with 48% in those in the steppedcare across attacks group and 29% in the steppedcare within attacks group. The most frequentlytaken rescue medication in the stratified, steppedcare across attacks and stepped care within attacksgroups, respectively, was zolmitriptan (58, 23 and66% of rescues) which was set by the protocol.Other medications used included naproxen (6, 11and 6%) and aspirin (6, 18 and 4%).

As regards adverse events considered severe,serious or which resulted in a patient’s withdrawalfrom a study, 10/279 (4%) patients in the stratifiedgroup experienced such events, compared with8/271 (3%) in the stepped care across attacks groupand 11/285 (4%) in the stepped care within attacksgroup. The most frequent actions estimated forthese adverse events were a visit to a general prac-titioner (GP) (100%), a visit to the accident andemergency department (34%) and a visit to a con-sultant neurologist (21%).

Unit Costs

The key unit costs used in the analysis were theacquisition cost of zolmitriptan (£4 per 2.5mg tab-let) and the cost of an hour of lost time based onthe average hourly wage rate in the UK (£10). Forcomparison, £1 = $US1.40. The acquisition costsof the rescue medications, other than zolmitriptan,were modest. The unit costs of the resources most

Table I. Key resource use per patient over six attacks

Item of resource use Stratified care Stepped care across attacks Stepped care within attacks

mean median (IQR) mean median (IQR) mean median (IQR)

Use of zolmitriptan (mg)a 15.42 15.00 (15.00, 20.00) 3.65 2.50 (0.00, 7.50) 10.88 10.00 (3.75, 15.00)

Lost work equivalents (h) 11.19 6.75 (0.00, 14.40) 14.43 9.83 (0.00, 19.16) 12.72 6.00 (1.25, 15.75)

a Including for rescue medication.

IQR = interquartile range.



frequently consumed as a result of adverse eventswere GP visit (£16), accident and emergency visit(£36) and visit to a consultant neurologist (£96).

Overall Costs

Table II shows the cost of managing patients inthe three arms of the trial over six attacks. The tableshows the cost of each category of resource use,including productivity costs.

Productivity costs represented the highest costcategory over 6 attacks, with mean per patient costsof £112.22, £144.70 and £127.53 for stratified care,stepped care across attacks and stepped care withinattacks, respectively. Compared with these costs,the cost of zolmitriptan was relatively small, andwas highest in the stratified group: mean £24.67versus £5.85 (stepped care across attacks) and£17.40 (stepped care within attacks). The absolutecosts of rescue medications other than zolmi-triptan, and of the management of adverse eventswere modest: the former ranged from £0.93 (strat-ified care) to £3.15 (stepped care within attacks)and the latter ranged from £2.60 (stepped carewithin attacks) to £3.91 (stepped care across at-tacks).

The mean health service cost per patient oversix attacks was highest in the stratified care group,followed by the stepped care within attacks group;the lowest health service costs were associatedwith stepped care across attacks. When productiv-ity costs were included, however, the ranking ofmean costs per patient changed. Stepped careacross attacks became the most costly (mean£157.19), followed by stepped care within attacks(mean £148.53) with stratified care being the leastcostly (mean £138.95). The overlapping boot-strapped confidence intervals shown in table II in-dicate, however, that total costs were not statisti-cally significantly different.

Effectiveness

Table III shows the primary effectiveness resultsin the trial. Both the mean 2-hour headache re-sponse over the 6 attacks and the mean disability-adjusted time over 6 attacks (maximum: 4 hours× 6 attacks = 24 hours) were higher in the stratifiedcare group than in the two stepped care groups. Thebootstrapped 95% confidence intervals in the tableshow that, in terms of headache response, stratifiedcare was statistically significantly more effectivethan both stepped care across attacks and stepped

Table II. Overall cost results per patient over six attacks (£; 1999 to 2000 prices)

Cost item Stratified care Stepped care across attacks Stepped care within attacks

mean median (IQR) mean median (IQR) mean median (IQR)

Use ofzolmitriptana

24.67 24.00 (24.00,32.00)

5.85 4.00 (0.00, 12.00) 17.40 16.00 (6.00,24.00)

Rescuemedicationsb

0.93 0.44 (0.43, 0.78) 1.98 0.89 (0.61, 1.39) 3.15 1.98 (1.98,2.97)

Adverseevents

2.65 0.00 (0.00, 0.00) 3.91 0.00 (0.00, 0.00) 2.60 0.00 (0.00,0.00)

Total healthservice costc

28.25 25.30 (24.43,36.43)

11.74 5.30 (1.14, 12.67) 23.15 21.98 (9.98,29.98)

Productivitycosts d

112.22 67.70 (0.00,144.43)

144.70 98.54 (0.00,192.20)

127.53 60.18 (12.54,157.97)

Total costs 138.95 (95% CI122.38, 158.93)

88.61 (32.43,170.85)

157.19 (95% CI134.50, 184.93)

91.19 (12.87,196.81)

148.53 (95% CI123.04, 179.41)

80.19 (29.93,173.38)

a Including as rescue medication.

b Other than zolmitriptan.

c Includes all drug costs and non-drug costs associated with serious or severe adverse events.

d Based on valuation of lost work equivalents at the average wage per hour.

CI = confidence interval; IQR = interquartile range.

96 Sculpher et al.


care within attacks. Using disability-adjusted time,stratified care was statistically significantly moreeffective than stepped care across attacks, but notagainst stepped care within attacks.

Given the nature of the management strategy, aproportion of patients in the stepped care acrossattacks group who received aspirin plus metoclo-pramide in attacks 1 to 3 stepped up to zolmi-triptan for attacks 4 to 6 because of poor responsein the early attacks (56, 69 and 74% of patientswith baseline MIDAS Grades II, III and IV, respec-tively, stepped up). As a result, the effectiveness ofstratified care and stepped care across attacks forattacks 4 to 6 were very similar. Hence, the statis-tically significant improved outcomes in the strat-ified care group compared with the stepped careacross attacks group across all six attacks waslargely a result of differences during attacks 1 to 3when migraineurs in the stepped care group werereceiving inappropriate treatment.

Cost Effectiveness

As regards the deterministic cost-effectivenessanalysis, based on mean costs (table II) and meaneffectiveness (table III) it can be seen that stratifiedcare dominates both stepped care across attacksand stepped care within attacks. This is becausestratified care has a lower mean health service plusproductivity cost and has a higher mean effective-ness, both in terms of response rate and disability-adjusted time.

Given the uncertainty around mean costs andeffectiveness (illustrated by the confidence inter-vals in tables II and III), it follows that there isuncertainty around conclusions about cost effec-tiveness. Figures 1 and 2 show two sets of cost-effectiveness acceptability curves, one for each ofthe two measures of effectiveness. These figurespresent the probability (y-axis) that one form ofmanagement is more cost effective than the othertwo for a series of maximum values (x-axis) thatdecision makers are willing to pay for an extra unitof effectiveness. The figures show that, from asocietal perspective including productivity costs,stratified care has the highest probability of beingcost effective whatever the maximum willingnessto pay per additional unit of effect. If decision mak-ers are willing to pay at least £1.20 per additional1% response rate and £20.00 per additional dis-ability-adjusted hour, the probability that stratifiedcare is more cost effective than the two steppedcare strategies reaches 90%. Decision makers willneed to consider their own maximum willingnessto pay for these outcomes in interpreting these re-sults. Even if they are not interested in improvinghealth outcomes (the value they place on improvedoutcomes is zero), the probability that stratifiedcare is cost effective (in this case cost saving) isover 50%.

Discussion

The DISC study has shown that a stratified carestrategy, based on zolmitriptan, is more effective

Table III. Results of the primary measures of effectiveness in the trial over six attacks

Effectivenessmeasure

Stratified care Stepped care across attacks Stepped care within attacks

mean (95% CI) median (IQR) mean (95% CI) median (IQR) mean (95% CI) median (IQR)

2-hour headacheresponse (%)

51.37 (47.30,55.68)

50.00 (16.67,83.33)

38.97 (35.14,42.84)

33.33 (0.00, 60.00) 36.48 (32.46,40.31)

33.33 (0.00,66.67)

Disability adjustedtime (h)a

12.90 (12.38,13.47)

13.14 (10.04,16.40)

11.43 (10.92,12.00)

11.50 (8.20, 14.68) 12.02 (11.43,12.56)

12.28 (8.54,15.33)

a Based on patients’ assessment of their level of functioning on a visual analogue scale, running from 0 (complete lack of function) to 100(complete function) at four time points before and after treatment. These responses were used to adjust the 4-hour post-treatment pe-riod for each attack. No disability over the four time points would result in 4 (4 x 100%) disability-adjusted hours; and complete disabil-ity over the four time points would result in 0 (4 x 0%) disability-adjusted hours. These results are presented over six attacks (i.e.maximum of 24 disability-adjusted hours).

CI = confidence interval; IQR = interquartile range.



than two forms of stepped care.[6] However, it wasalso necessary to demonstrate that this strategy rep-resents a cost-effective use of resources. The anal-ysis reported here has shown that, on the basis ofmean (health service plus productivity) costs, strat-ified care has a lower mean cost per patient and hasa higher mean level of effectiveness, whether thisis measured in terms of 2-hour response rates ordisability-adjusted time over 4 hours. In otherwords, a stratified care strategy which includes zol-mitriptan for the acute treatment of migraine is the‘dominant’ treatment strategy, and is unequivo-cally more cost effective than the two forms ofstepped care.

Despite the size of the trial (835 patients in theefficacy analysis), the precision of the estimates ofmean cost and effectiveness needs to be consid-ered. One potentially valuable approach is to in-form decisions makers about the probability thatstratified care will be more cost effective than thetwo forms of stepped care given data from theDISC study. Figures 1 and 2 show that this prob-ability ranges from about 60% to just below 100%

for both measures of outcome, depending on theamount decision makers are willing to pay for anadditional unit of effect for patients. Even if deci-sion makers are not willing to pay anything addi-tional for improved benefits for patients, the prob-ability that stratified care saves resources isapproximately 60%.

These probabilities (which are Bayesian in theirinterpretation[26]) are lower than would usually beused to reject a null hypothesis of no difference incost effectiveness between the alternative strate-gies. That is, the cost effectiveness of stratified careis only ‘statistically significantly’ more cost effec-tive (using a conventional p value of 0.05) than thetwo forms of stepped care when decision makersplace a high value on an outcome (and neveragainst stepped care within attacks when disabil-ity-adjusted time is the measure of outcome). How-ever, in deciding which interventions representgood value for money, there is good reason not toadhere to standard methods of statistical infer-ence;[27] if decision makers wish to maximisehealth gain from finite resources, they should base

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0 1 2 3 4 5 6 7 8 9 10

Pro

babi

lity

cost

effe

ctiv

e

Value (£) for extra 1% response rate

Stratified careStepped care across attacksStepped care within attacks

Fig. 1. Cost-effectiveness acceptability curves using 2-hour re-sponse rate as the measure of effectiveness. The x-axis pres-ents the maximum value decision makers attach to an additional1% response rate. The y-axis shows the probability that eachof the three treatment strategies is more cost effective than theother two.

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0 10 20 30 40 50 60 70 80 90 100

Pro

babi

lity

cost

effe

ctiv

e

Value (£) for extra disability-adjusted hour

Stratified careStepped care across attacksStepped care within attacks

Fig. 2. Cost-effectiveness acceptability curves using disability-adjusted time as the measure of effectiveness. The x-axis pres-ents the maximum value decision makers attach to an additionaldisability-adjusted hour. The y-axis shows the probability thateach of the three treatment strategies is more cost effective thanthe other two.

98 Sculpher et al.


decisions on expected (mean) costs and outcomes.If standard statistical tests are used with arbitraryp-values as decision thresholds, this amounts to adecision that standard management (presumably aform of stepped care) should be used althoughthere is between a 60 and over 90% chance of thewrong decision being taken. Taken together, thedata presented in this paper would suggest thatdecision makers have the greatest chance of max-imising patients’ health from limited (societal)resources through the use of stratified care.

How generalisable are these results? The firstpoint to make is that patients in the trial did notinclude those with MIDAS Grade I migraine, andpatients with mild migraine were encouraged notto treat their migraines using trial medications.These details should be considered when interpre-ting the results of the study in the context of thegeneral population of patients with migraine.

Secondly, although the costs presented here arepresented in UK currency, the DISC trial includedpatients from 13 countries. Due to variability inclinical practice between countries, this raises thequestion of whether the costs and effects estimatedin the trial are generalisable to routine practice inthe UK. Given the nature of the strategies undercomparison, which centre around the timing ofthe use of specific drugs and, for most patients, donot include any other tests or therapeutic interven-tions, it is unlikely that effectiveness and resourceuse would differ markedly between countries. Theonly area of resource use that may vary betweendifferent systems is the management of adverseevents where, for example, hospital visits for agiven adverse event may be more likely in somecountries than others. However, the cost of manag-ing adverse events was a small proportion of theoverall costs of care (ranging from 1.6 to 2.4%), sothis would not affect the conclusions of the analy-sis.

The conclusions presented here do depend onthe adoption of a societal perspective to the analy-sis. The majority of formal guidance for economicevaluation methods to inform health service deci-sion making include a role for productivity costs.

These include guidance offered by the US Panel onCost-Effectiveness in Health and Medicine,[15] theCanadian Co-ordinating Office for Health Tech-nology Assessment,[28] and by the Australian[29]

and Dutch[30] governments. To exclude productiv-ity costs from assessments of the cost effectivenessof alternative interventions which differentiallyaffect patients’ time away from paid work is toignore an important resource implication of careand to impose costs on other sectors of the econ-omy simply to protect health service budgets.[31]

Conclusion

Given its lower mean costs (health service plusproductivity) and higher mean effectiveness, astratified care strategy including zolmitriptan wasthe dominant strategy for the acute treatment ofmigraine. Allowing for the uncertainty in thesemean values, stratified care had the highest prob-ability of being cost effective.

Acknowledgements

The authors would like to thank Dr Anne MacGregorfrom the City of London Migraine Clinic for her helpfuldiscussions relating to the clinical scenarios associated withmanaging the adverse events. The DISC trial and this eco-nomic evaluation were funded by AstraZeneca.

References1. Saper JR, Silberstein S, Gordon CD, et al., editors. Handbook

of headache management: a practical guide to diagnosis andtreatment of head, neck and facial pain. Baltimore (MD): Wil-liams and Wilkins, 1993

2. Diener HC, Kaube H, Limmroth V. A practical guide to themanagement and prevention of migraine. Drugs 1998; 56:811-24

3. Pryse-Phillips WE, Dodick DW, Edmeads JG. Guidelines forthe diagnosis and management of migraine in clinical prac-tice. Canadian Headache Society. CMAJ 1997; 156: 1273-87

4. Lipton RB. Disability assessment as a basis for stratified care.Cephalalgia 1998; 18 Suppl. 22: 40-6

5. Lipton RB, Stewart WF. Clinical applications of zolmitriptan(Zomig, 311C90). Cephalalgia 1997; 18: 530-9

6. Lipton R, Stewart WF, Stone A, et al. Stratified care vs stepcare strategies for migraine: the Disability in Strategies ofCare (DISC) study: a randomised trial. JAMA 2000; 284:2599-605

7. Williams P, Dowson AJ, Rapoport AM, et al. The cost effective-ness of stratified care in the management of migraine. Phar-macoeconomics 2001; 19: 819-29

8. Stewart WF, Lipton RB, Whyte J. An international study toassess reliability of the Migraine Disability Assessment(MIDAS) score. Neurology 1999; 22: 988-94



9. Stewart WF, Lipton RB, Kolodner K. Reliability of the Mi-graine Disability Assessment Score in a population-basedsample of headache sufferers. Cephalalgia 1999; 19: 107-14

10. Stewart WF, Lipton RB, Kolodner KB. Validity of the MigraineDisability Assessment (MIDAS) score in comparison to adiary-based measure in a population sample of migraine suf-ferers. Pain. In press

11. British Medical Association, Royal Pharmaceutical Society ofGreat Britain. British National Formulary. London: BritishMedical Association and the Royal Pharmaceutical Society ofGreat Britain, no. 39, 2000 Mar

12. Netten A, Dennett J, Knight J. Unit costs of health and socialcare. Canterbury: Personal Social Services Research Unit,University of Kent, 1999

13. Chartered Institute of Public Finance and Accountancy(CIPFA). The Health Service Database 1999. Croydon:CIPFA, 1999

14. MEDTAP International. Database of International Unit Costsfor Economic Evaluation in Health Care. London: MEDTAPInternational Inc, 1999

15. Gold MR, Siegel JE, Russell LB, et al., editors. Cost-effective-ness in health and medicine. New York (NY): Oxford Univer-sity Press, 1996

16. Brouwer WBF, Koopmanschap MA, Rutten FFH. Productivitycosts in cost-effectiveness analysis: numerator or denomina-tor: a further discussion. Health Econ 1997; 6: 511-4

17. Office for National Statistics. Monthly digest of statistics, April.London: The Stationary Office, 2000

18. Koopmanschap MA, Rutten FFH, van Ineveld BM, et al. Thefriction cost method of measuring the indirect costs of disease.J Health Econ 1995; 14: 123-262

19. Efron B, Tibshirani R. An introduction to the bootstrap. NewYork (NY): Chapman & Hall, 1993

20. Johannesson M, Weinstein S. On the decision rules of cost-ef-fectiveness analysis. J Health Econ 1993; 12: 459-67

21. Van Hout BA, Al MJ, Gordon GS, et al. Costs, effects and c/e-ratios alongside a clinical trial. Health Econ 1994; 3: 309-19

22. Fenwick E, Claxton K, Sculpher M. Representing uncertainty:the role of cost-effectiveness acceptability curves. HealthEcon 2001; 10: 779-89

23. UK Prospective Diabetes Study Group. Cost effectiveness anal-ysis of improved blood pressure control in hypertensive pa-tients with type 2 diabetes: UKPDS 40. BMJ 1998; 317: 720-6

24. Delaney BC, Wilson S, Roalfe A, et al. Cost effectiveness ofinitial endoscopy for dyspepsia in patients over age 50 years:a randomised controlled trial in primary care. Lancet 2000;356: 1965-9

25. Briggs AH, Gray A. Handling uncertainty when performingeconomic evaluation of healthcare interventions. HealthTechnol Assess 1999; 3: 1-134

26. Briggs AH. A Bayesian approach to stochastic cost-effective-ness analysis. Health Econ 1999; 8: 257-62

27. Claxton K. The irrelevance of inference: a decision-making ap-proach to the stochastic evaluation of health care technolo-gies. J Health Econ 1999; 18: 342-64

28. Canadian Coordinating Centre for Health Technology Assess-ment (CCHOTA). Guidelines for Economic Evaluation ofPharmaceuticals: Canada. Ottawa: 1997

29. Commonwealth of Australia. Guidelines for the PharmaceuticalIndustry on Preparation of Submissions to the PharmaceuticalBenefits Advisory Committee: Including Major SubmissionsInvolving Economic Analyses. Canberra: Australian Govern-ment Publication Service, 1995

30. Health Insurance Council. Dutch guidelines for pharmaco-economic research. Amsterdam: Health Insurance Council,1999

31. Johannesson M, O’Conor RM. Cost-utility analysis from a so-cietal perspective. Health Policy 1997; 39: 241-53

Correspondence and offprints: Professor Mark Sculpher,Centre for Health Economics, University of York, ,Heslington, York, YO10 5DD, United Kingdom.

100 Sculpher et al.


Documents

Cost-Effectiveness Analysis of Stratified Versus Stepped Care Strategies for Acute Treatment of Migraine