conjoint analysis of a new chemotherapy

Conjoint Analysis of a NewChemotherapyWillingness to Pay and Preference for the Features of Raltitrexed versus Standard Therapy in Advanced Colorectal Cancer

Mike Aristides,1 Jack Chen,1 Mark Schulz,1 Eve Williamson,2 Stephen Clarke3 and Kaye Grant2

1 M-TAG Pty Ltd, Sydney, New South Wales, Australia2 AstraZeneca Pharmaceuticals, Sydney, New South Wales, Australia3 Royal Prince Alfred Hospital, Sydney, New South Wales, Australia

Abstract Objective: To estimate the willingness to pay for a new chemotherapy, raltitrexed(Tomudex™1) over conventional therapy with fluorouracil plus leucovorin(FU-LV) from the perspective of patients with advanced colorectal cancer. Thestudy was part of the product’s reimbursement application in Australia.

Design and Methods: The key differences relevant to patients between the twotherapies, frequency of administration and severity of mouth ulceration, wereused to develop a self-administered questionnaire. A relatively new technique tohealthcare, that of conjoint analysis (CA), was used to estimate willingness topay and strength of preference. A discrete choice CA was used. Analysis was viaa logit model with adjustment for the cluster effect (or intra-patient correlation).

Study participants: Oncology nurses served as proxies for patients with ad-vanced colorectal cancer.

Results: The participation rate was 87% (62/71) with questionnaires from 56respondents eligible for analysis. The CA method generated a mean incrementalwillingness to pay of 745 Australian dollars ($A; $US507) per cycle of chemo-therapy, comprising $A550 ($US374) and $A195 ($US133) for the toxicity andadministration improvements, respectively (1998 values). Both features wererelated to preference with independent odds of 6.87 and 1.98, respectively, howeveronly the toxicity attribute was a significantly related to preference. Subscriptionto private health insurance was the only significant demographic predictor iden-tified, however, the homogeneous income structure of the respondents seemslikely to have masked any significant income effect.

Conclusions: This study highlights the advantages of using CA in healthcare,where new therapies or treatments are often composed of a number of attributes.

ORIGINAL RESEARCH ARTICLE

1 The use of the trade name is for product identification purposes only and does not imply endorsement.

Pharmacoeconomics 2002; 20 (11): 775-7841170-7690/02/0011-0775/$25.00/0

© Adis International Limited. All rights reserved.

The CA allows both strength of preference and willingness to pay for individualtreatment attributes to be estimated and can be used to assign statistical signifi-cance to these parameters.

The purpose of deriving a patients willingnessto pay for healthcare services is to value the rela-tive benefits which a therapy offers.[1] This valuecan be used in an economic evaluation of a newintervention in order to help decision makers de-termine whether the new intervention should beadopted or rejected.

Two major approaches used to determine will-ingness to pay are contingent valuation (CV) orconjoint analysis (CA). Contingent valuation isnormally carried out using a direct survey of con-sumers and asks what is the maximum the con-sumer is willing to pay for a particular health ben-efit. The CV approach has been used to value avariety of healthcare programmes or the health out-comes which may be expected from a particulartherapy.[2-8] A relatively new alternative to deter-mining willingness to pay is that of CA. This tech-nique stems from the market research area and wasinitially applied to determine the strength of pref-erence for individual features of products.[9] Morerecently, the technique has been applied in healtheconomics to assess the relative importance indi-viduals place on the attributes of health interven-tions and their willingness to pay for them.[10-14]

Ryan and Farrar[15] provide a comprehensive re-view of its use in the healthcare area. In discretechoice CA, respondents are asked to make a num-ber of pair-wise choices comparing scenarios thatcontain combinations of attributes. These compar-isons are designed to elicit trade-offs from the re-spondent in order to determine their preferences.

Since 1993, an economic evaluation has beenrequired in Australia before a pharmaceutical canbe subsidised on the national formulary. The eval-uations are for consideration by the national for-mulary committee and a technical sub-committeewithin the Federal Department of Health.[16] Eval-uating chemotherapeutic agents may at times bedifficult particularly where clinical efficacy is sim-ilar but differences may be evident in the adverse

effects they cause. Clearly, individuals may givegreater value to treatments where adverse effectsoccur at a lower rate or are absent. As part of aneconomic evaluation for the Australian nationaldrug formulary, a willingness-to-pay (WTP) studywas conducted comparing a new chemotherapyraltitrexed (Tomudex™) versus fluorouracil pluslow dose calcium folinate (FU-LV) which is con-sidered standard therapy in Australia for advancedcolorectal cancer.

Raltitrexed is a specific inhibitor of thymidylatesynthase, an enzyme that plays an important rolein DNA synthesis. This inhibitory activity makesraltitrexed an effective and specific chemothera-peutic agent and its greatest activity has been ob-served in patients with advanced colorectal cancer.Although FU-LV is considered standard treatmentfor this type of cancer, the use of modified ana-logues of DNA such as fluorouracil, although ef-fectively inhibiting cancer growth, are rather non-specific and often result in numerous adverseeffects during treatment. In a number of clinicaltrials, the incidence of hair loss, mouth ulcerationand other oral effects were lower in patients treatedwith raltitrexed compared to FU-LV.[17,18] Ral-titrexed also offers advantages over FU-LV in themethod of administration, with raltitrexed admin-istered 1 day every 3 weeks and FU-LV adminis-tered for 5 consecutive days every 4 weeks.

This study aimed to compare the willingness topay for chemotherapy with raltitrexed comparedwith the standard treatment, FU-LV. Differences inthe administration and adverse effects observedduring clinical trials were used to develop scenar-ios comparing the two treatments. Scenarios incor-porating different levels of administration, toxicityand cost were presented to elicit trade-offs betweenthese attributes. The scenarios were then used todetermine the willingness to pay for a particulartreatment.

776 Aristides et al.

© Adis International Limited. All rights reserved. Pharmacoeconomics 2002; 20 (11)

Methods

Study Sample

Patients with advanced colorectal cancer werenot considered suitable respondents in this studydue to severity of disease and the likely stresscaused in conveying or confirming their poor prog-nosis. In similar situations, healthcare workershave been used as patient proxies because they arefamiliar with the problems and adverse effects re-sulting from a particular treatment.[19] For suchreasons, we investigated whether oncology nursesmay make suitable proxies for this study. A pilotWTP study of oncology nurses (n = 10) was carriedout using face-to-face interviews. The oncologynurses understood the questions, and were familiarwith the various attributes of chemotherapy. Theydid not object to the approach used and all (100%)traded rationally in the interview. On this basisthey were considered suitable respondents for afull study.

Four oncology clinics in large Sydney teachinghospitals agreed to participate and a total of 71nurses, all with experience in oncology and che-motherapy administration, were enrolled. The entiresurvey was self-administered by the respondent. Itcontained two sections concerning demographicdetails and the CA, respectively.

Treatment Differences

The alternate treatment characteristics were de-rived from the results of two pivotal phase III trialsof raltitrexed which assessed relevant compara-tor regimens for Australia.[17,18] Specifically, thetreatments were distinguished by the incidence ofclinically significant toxicity (mucositis and otheroral effects), and the frequency of chemotherapyadministration.

The incidence of oral effects, as defined by stand-ard World Health Organization (WHO) grades,[19]

was pooled from the two-phase III clinical trials.The mean frequency of all grades (I-IV) and severegrades (III/IV) of mucositis was determined. Theseare outlined in table I. Lower grades (I/II) ofmucositis are characterised by a sore and dry mouth

often with a burning sensation. Severe disease(grade III/IV) is characterised by extensive mouthulceration that makes it impossible to eat any solidfood.

Raltitrexed and FU-LV both involve adminis-tration by infusion in the outpatient setting. How-ever, raltitrexed involves only one treatment every3 weeks compared to treatment on 5 consecutivedays every 4 weeks with FU-LV. These scheduleswere used to represent a cycle of treatment for eachchemotherapy (see table I).

It should be noted that during clinical trials asignificantly lower amount of grade I & II alopecia(hair loss) was observed in patients receivingraltitrexed compared to FU-LV. However, this fea-ture was not found to influence a respondent’swillingness to pay during the pilot study and wastherefore omitted.

Conjoint Analysis

This study used discrete-choice CA. Individu-als were asked to choose between two treatmentoptions. The profiles of the two treatments werepresented in a full profile format, where treatmentswere thoroughly described in terms of all attri-butes. An example of a choice comparison is pro-vided in table II.

Levels of AttributesThe three key attributes evaluated were risk of

mucositis (using both WHO grade III/IV mucositisand any grade of mucositis), administration (usingdays administered over the number of weeks; seetable I for details) and cost. Attribute levels were

Table I. Differences in mucositis and administration during treatmentwith raltitrexed and fluorouracil plus low dose calcium folinate (FU-LV)

Raltitrexed FU-LV Difference

Mucositis (%)All grades 19 63 44 (95% CI: 38-50)Grade III/IV 2 19 17 (95% CI: 13-21)

AdministrationDaysa/weeks perchemotherapy cycle

1/3 5/4 11b/12

a Consecutive days of treatment.b 15 days minus 4 days.CI = confidence interval.

Conjoint Analysis of a New Chemotherapy 777


assigned such that there were four levels of toxic-ity, four levels of days spent in administration andseven levels of cost (i.e., 112 possible combina-tions). Of the four levels of mucositis and admin-istration included, two were derived from the clinicalstudies while two additional levels were includedto better examine the preferences of individuals forthese attributes. The cost attribute included sevenlevels that covered the range of 0 to 800 Australiandollars ($A; 1998 values).

The 112 possible scenarios were manually ex-amined to ensure that illogical pair-wise choiceswere eliminated. The majority were removed be-cause the baseline scenario dominated or was bet-ter than the alternate scenario at no cost. Choiceswhere marginal benefits had a large cost were alsoremoved. After this review of the 112 possible sce-narios, 73 were eliminated on the criteria outlinedabove. This left a total of 39 possible pair-wisechoices that were used in the questionnaire. Thisset provided a wide range of attributes and cost forassessment. The schedules of attributes and costlevels are presented in table III and table IV.

During the study the costs were framed in termsof an additional ‘out-of-pocket’ cost to the patientwhereas the baseline scenario was presented ashaving no cost. An example of a question used inthe study is presented in table II.

Rational Trading and Lexicographic EffectIn the present study, two ‘dominated’ pair-wise

choices were included. In these, one of the treat-ments was superior in all attributes and levels.These were designed to detect so-called irrationaltraders or respondents who choose a scenario even-though all the attributes of the opposing scenario

are superior. Additionally, individuals whose choicedecision was based solely on zero cost were re-moved from the analysis. Such individuals weredefined as those where they made a no cost choicein all scenarios and was based on the assumptionthat the random set should have contained at leastone scenario where they were offered superior at-tribute levels at a reasonable cost level.

Allocation of Questionnaire to ParticipantsExtensive work has examined how best to de-

rive an efficient set of pair-wise choices for respon-dents (see, for example, Greene[20]). An orthogonalarray is an experimental design which assumesaway most (sometimes all) interactions among theindependent variables. In other words, the effectsof each selected attribute and attribute level arewell balanced and kept separate from those of an-other.[20] In most cases such work has been con-ducted in more general consumer products and notin health. We used an approach that randomly as-signs sets of the possible scenarios to each individ-ual. This process is appropriate where the interac-tions between attributes and levels is not clear, asis the case in the attributes used in this study. Weensured minimal overlap by randomising using atwo-step approach, based initially on the mucositisand administration attributes and levels and thenon cost. This ensured that individuals did not re-ceive the same treatment choices at varying costlevels. This approach was adopted because thestudy aimed to examine choice in the entire popu-lation and not at an individual level.

To avoid potential bias in scenario selection andordering, the 39 scenarios were assigned a randomnumber generated using a uniform distribution

Table II. Example of conjoint analysis question

AdministrationOne cycle of treatment involves a single injection: For 8 consecutive

days every 28 daysFor 5 consecutivedays every 28 days

Mouth ulcerationEvery time the drug is administered there is a chance that your mouth will become quitesore and dry and you will experience a burning sensation

19% 63%

There is also a possibility that mouth ulceration will make it impossible for you to eat anysolid food

2% 19%

Cost to you of each cycle (1998 Australian dollars; $A) $A200 $A0Which option do you prefer? (Please tick one box) Option A Option B



function. Sets of scenarios and their order werethen derived randomly. Respondents were giveneither 10 or 15 pair-wise choices to determinewhether any difference in responses were evidentif respondents received more scenarios. To exam-ine this issue, one of the two ‘dominated’ pair-wisechoices, mentioned previously, was inserted at theend of the 10 and 15 choice sets. The hypothesiswas that if the larger set caused cognitive problemsthen more respondents would have traded irratio-nally after receiving 15 sets than 10. No significantdifferences were observed in the number of irratio-nal traders between the two sets.

Framing of Willingness-to-Pay Questions

The scenarios start with a brief outline of thecondition (colorectal cancer) which the respondentimagines themselves suffering. The respondentswere also asked to imagine that their life expec-tancy was approximately 10 months (based on me-dian survival times observed in clinical trials com-paring raltitrexed and FU-LV).[17,18]

Mucositis was framed as a population risk thatmay occur each time treatment was received. Asthe study respondents were nurses this conceptshould have been familiar to them. During the pilotstudy all indicated that they understood the de-scription used.

Statistical Analysis

GeneralData was entered by one professional data entry

clerk and double-checked by randomly sampling

5% of the original data. Range and logic checkingwere carried out to ensure the quality of the database.

A total of 735 scenarios were included in thepreliminary database. A set of eligible question-naires was identified based on the exclusion of re-spondents with lexicographic preference, irratio-nal trading or incomplete questionnaires.

Descriptive and Bivariate AnalysisDescriptive statistics were performed on all of

the responses collected during the interview. Forcategorical responses, the number of respondentsand the percentage of respondents falling into eachcategory were assessed.

Conjoint AnalysisA logit regression model was used to regress the

stated preference for the pairs of scenarios againstthe levels of toxicity, days in administration andcost, as well as potential confounding variablessuch as personal income, gender, private health in-surance status and age.[20] Specifically, the robustcluster logit model was used[21,22] to allow for thenon-independence of the observations or ‘clustereffect’ (i.e., multiple records from the same re-spondent). This method of dealing with clustereddata has a long history in survey statistical litera-ture[23,24] and has been previously used in anothereconomic application of CA.[25]

The logit model estimates the dependent vari-able logit (Li) which is defined as the natural logof the odds of preferring a scenario (i.e., ln[Pi/1-PI]). A key feature of such models is that the de-pendent variable is a linear function of the inde-pendent variables and the coefficients.[26] Themodel specification is in line with additive utilitytheory.[27]

The general model is specified as follows(equation 1):

where the equation is defined as follows: P = prob-ability of an individual choosing a tradable sce-nario j [j = 1…n]; α = model constant; Ai = admin-istration characteristics [four levels, three dummy

Table III. Attributes and levels used in the conjoint analysis

Attribute Levels Description

Mucositis 4 Risk of both WHO CTC Grade III/IVand WHO CTC Grade I-IV mucositis

Administration 4 Number of administrations requiredper chemotherapy cycle

Cost (Australiandollars; $A;1998 values)

7 Cost of treatment (range used was$A0-$A800)

WHO CTC = World Health Organization common toxicity criteria.[19]

U =

ln(P

1P ). = + A + T + C + D +

jj ij ij ij ij ij ij i i− α β γ δ λ ε



variables, i = 1,2,3]; Ti = toxicity characteristics[four levels, three dummy variables, i = 1,2,3]; Ci

= cost values of treatment [seven levels, $A0-800,treated as a continuous variable]; Di = importantdemographic variables [age and a vector of dummyvariables including income, private health insur-ance status and gender]; ε = error term.

A feature of the discrete choice design in thisstudy is that a fixed scenario was used (FU-LV)and all other tradable scenarios were comparedwith the fixed scenario. The value of ln[p/(1 - p)]can be defined as the incremental utility score of ascenario. If any scenario is preferable to FU-LV,the probability of this scenario being chosen mustbe greater than 50% and the utility value of thisscenario is larger than zero. If a scenario is deemedas equal to FU-LV, the probability of this scenarioto be chosen will be 50% and the incremental util-ity is zero (i.e., ln[0.5/1-0.5] = 0). Hence, the base-line scenario’s utility score is implicitly set to zero.This is because the baseline scenarios are set asreference categories. Determining whether themodel constant is significantly different from zerowill test this assumption.

The coefficients β, γ or δ in equation 1, reflectthe change in utility for each attribute using theindividuals stated preference (holding other attri-butes and income constant). As demonstrated pre-viously by other researchers, the ratio of any twoof the model coefficients indicates the individualsmarginal rate of substitution between the attri-butes.[15,27,28] In particular, a ratio including thecost coefficient δ indicates the willingness to payfor changes in a unit of the other attribute (e.g. the

marginal rate of substitution of higher cost for alower category of toxicity).

The overall willingness to pay for raltitrexed overFU-LV is estimated by the willingness to pay forthe improvement in toxicity plus the willingness topay for the improvement in the burden of adminis-tration.[28,29] In addition, the model coefficients andwillingness to pay for each attribute level is reported.The modelling process was divided into two steps.

First, a logit model was developed which usedadministration, toxicity and cost as the only predic-tors (model 1). The relative preference for eachattribute level was determined, as was the willing-ness to pay. The overall willingness to pay for aparticular therapy was calculated by summing theindividual WTP values for each attribute level.

Second, possible confounding factors (i.e., in-come, age, gender and private health insurance sta-tus) were incorporated into the logit model (model2). The main effects (i.e., attributes of treatment)were first examined in the model, while the inter-action effects between income and each attributewere estimated using a backward stepwise method.Model 2 therefore provides an adjustment for pos-sible confounding factors. The significance levelwas set at the 0.01 level for the test for interactioneffects to avoid chance results, and at the 0.05 levelfor all other tests. STATA™ 6.0[30] software wasused for the analyses.

Results

Demographics

The demographic information obtained duringthe study is summarised in table V. Eighty-seven

Table IV. Levels of mucositis and administration used in the conjoint analysis

Levels1 2 3 4 5 6 7

Mucositis (%)All grades 75 63 45 19Grade III/IV 30 19 10 2

AdministrationDaysa/weeks per chemotherapy cycle 8/4 5/4 2/3 1/3Cost (Australian dollars; 1998 values) 0 100 200 300 400 600 800a Consecutive days of treatment.



per cent (62/71) completed the self-administeredWTP questionnaires, resulting in a total of 735 sce-narios. Respondents were evenly represented acrossstudy centres. The ages of the nurses ranged from 21to 57 years and were relatively evenly distributed.Seventy-nine per cent specialised in oncology. Themajority of nurses in the survey were female(88.7%), employed full-time (72.6%) and married(71.0%). Annual family incomes ranged in thebrackets of $A20 000 to $A35 000 ($US13 600 to$US23 800) to >$A100 000 ($US68 000), withonly 9.7% having incomes above $A100 000. Ap-proximately half (54.4%) were privately insured.When respondents were directly offered a choicebetween treatments based on FU-LV and ralti-trexed with no additional cost, all respondents(100%) chose raltitrexed.

Conjoint Analysis

Rational Trading and Lexicographic EffectsAfter random allocation, 30 nurses received one

of the dominated scenarios. Almost all respon-dents (28/30) made a rational choice for the base-line scenario. These results indicate that the major-ity of respondents traded in accordance with the

theoretical expectation. The two respondents whodid not make a rational choice were excluded fromthe final analysis.

Of the remaining 60 respondents who com-pleted the questionnaire, four respondents demon-strated a lexicographic response and these respon-dents were also excluded. In total 658 scenarioswere valid for analysis.

Model ValidityThe results derived from the Hosmer and

Lemeshow Goodness-of-Fit Test was 5.19 (df = 8;p = 0.74) indicating that the model is a reasonablefit of the data. The model coefficients presented intable VI can be interpreted as the change in utilityscore between levels of the attributes. The sign ofthe coefficients support the theoretical expectationthat worse levels of toxicity and administration com-pared to baseline have a negative sign and betterlevels have a positive sign. The sign for the costattribute is negative indicating less utility withmore cost.

Logit Regression ModelThe estimated coefficients for model 1 are pre-

sented in table VI. Cost and toxicity were highlysignificant predictors of choice (p < 0.001) whilethere were trends for the administration scheme.

Keeping other features equal, improving themucositis level from that of FU-LV (gradeIII/IV:19%; grade I-IV: 63%) to that of raltitrexed(grade III/IV: 2%; grade I-IV: 19%) increased theutility score by 1.930, and respondents were 6.87(95% CI: 3.27 to 14.45) times more likely to choosethis improved scenario. The willingness to pay forthis improvement was $A550 ($US374). Changingthe administration scheme from that of FU-LV (5days every 4 weeks) to that of raltitrexed (1 dayevery 3 weeks) yielded a 0.685 increase in utilityscore, and respondents were 1.98 times more likelyto favour this change. Their willingness to pay forthis change was $A195 ($US133). Overall, the at-tributes of raltitrexed increased the utility score by2.615 (1.930 + 0.685), with an estimated WTP payper cycle of $A745 ($US507) for both improve-ments or $A550 when only significant predictorsof choice (mucositis) are included.

Table V. Characteristics of study population (n = 62)

Characteristic Value

Mean age in years (range) 36.8 (21-57)

Female (%) 88.7

Nursing specialty (%)Oncology 79.0

Other 21.0

Marital status (%)Married/de facto 71.0

Single 25.8

Divorced 3.2

Incomea (%)$A20 000-$A35 000 ($US13 600-$US23 800) 21.0

$A35 000-$A50 000 ($US23 800-$US34 000) 30.6

$A50 000-$A100 000 ($US34 000-$US68 000) 38.7

>$A100 000 ($US68 000) 9.7

Private health insurance (%) 54.8

a Exchange rate $A/$US0.68 [approximate average value at thetime this study was undertaken (July, 1998)].

$A = Australian dollars.



In model 2, no evidence of interaction effectswas evident and only subscription to private insur-ance was significantly related to willingness topay. When the main effects of the demographiccharacteristics were included in model 2, the re-sults were not significantly different to those ob-tained using model 1.

Influence of Number of ScenariosTo examine whether cognitive problems may

have occurred when respondents were given morescenarios, an additional analysis was undertaken.The null hypothesis was the choice patterns of re-spondents given a random set of 10 compared with15 scenarios was equivalent. The choice patternsof respondents who received a random set of 10scenarios were not significantly different fromthose who received 15 scenarios (p = 1.0; FishersExact test). This indicates that both groups wereusing similar cognitive processes to make choicedecisions.

Discussion

This study was conducted to determine the will-ingness to pay for the features offered by a novelchemotherapy (raltitrexed) compared to standard

treatment (FU-LV) for use in advanced colorectalcancer. CA was used to estimate willingness topay. The study was part of an economic evaluationconducted to support the government subsidy of ralti-trexed on the Australian national drug formulary.Hence, these valuation techniques were used withina policy framework where economic data are re-quired and rigourously appraised.[31,32]

The application to reimburse raltitrexed on theAustralian national drug formulary was successful.Given the confidential nature of this process be-tween sponsor and funder, it is not possible to com-ment directly on the bearing of this study. How-ever, this study clearly contributed to the outcomesand economic assessments of the product providedby the sponsor. A forthcoming revision to the Aus-tralian pharmacoeconomic guidelines is expectedto address the use and techniques of willingness topay in this context.

The CA method elicits preference indirectly,uses multiple observations and provides a rich setof information such as the preference for attributes(in terms of probabilities or odds). Further, the CAmethod allows testing of theoretical and statisticalvalidity, including whether the model truly reflects

Table VI. Estimated utility equation coefficient, odds of preference and willingness to pay (WTP) of the respondents for the improvement ofadministration scheme and toxicity level (model 1; n = 658 scenarios: 56 respondents)

Variable Coefficient P value Odds ratio (95% CI) WTPa $A ($USb)

Administration (daysc/weeks per chemotherapy cycle)5/4d

8/4 –0.062 0.854 0.94 (0.49-1.72) –18 (12)

1/3 0.685 0.091 1.98 (0.90-4.39) 195 (133)

2/3 0.085 0.867 1.09 (0.40-2.95) 24 (16)

Cost –0.351* 0.000 0.70 (0.64-0.77)

Mucositis (% grade III or IV/% all grades)19/63d

2/19 1.930* 0.000 6.87 (3.27-14.45) 550 (374)

10/45 0.515 0.233 1.67 (0.72-3.90) 147 (100)

30/75 –1.452* 0.000 0.23 (0.13-0.42) 414 (282)

Constant 0.329 0.473

a Willingness to pay (marginal rate of substitution with cost).

b Exchange rate $A/$US0.68 [approximate average value at the time this study was undertaken (July 1998)].

c Consecutive days of treatment.

d Reference category.

$A = Australian dollars; CI = confidence interval; * p < 0.001.



incremental utility, the extent of rational economicbehaviour and extent of lexicographic behaviour.Therefore the CA is an appealing method for esti-mating willingness to pay alongside respondentpreferences.

Although there is some debate over the appro-priate group to provide values in such studies, util-ity and WTP values have previously been soughtfrom patients in Australia as advised by repre-sentatives of the national formulary committee.[33]

Although there have been no formal guidelines onthis matter in Australia, the social decision makerhas seen patient’s values as highly relevant. In thisstudy, nurses were used as proxy respondents forpatients who are logistically and ethically difficultto study directly. It is noted that these nurses haveexperience in the treatment of toxicity with chemo-therapy, whereas any individual patient may ormay not have experienced the differentiating ad-verse event. As such, there may be advantages inthe information/experience set that these nursesbring. Further, nurses’ incomes approximate thenational average. To the extent that their income issimilar to the patient group, these results can beseen as generalisable. Notably, there was no strongindication of high income elasticity.

As with all research, it would be preferable ifthe results were replicated. Study of patients directlywould add to these findings.

Although both toxicity and administration weresignificantly related to preference, toxicity was themost important feature. The improvements offeredby raltitrexed in the extent and severity of mucos-itis increased the utility score considerably and re-spondents were approximately seven times morelikely to choose a scenario with this level of toxic-ity. In this study, income was not significantly re-lated to willingness to pay and only subscription toprivate health insurance was related to choicebehaviour. This may be due to the relatively lim-ited spread of household incomes inherent to thisgroup.

In this study, cognitive overload was not evi-dent for those who completed 15 scenarios com-pared with 10. These results suggest that the re-

spondents could comprehend and continue tomake rational choices beyond 10 questions. Giventhe fact that 15 scenarios produces 50% more in-formation than 10 scenarios, future studies of sim-ilar design should consider carefully the number ofscenarios used.

Conclusions

In conclusion, the CA technique provided im-portant preference and WTP information. Respon-dents were willing to pay $A745 ($US507) per cy-cle of chemotherapy for raltitrexed over FU-LVwhen all attributes are included or $A550 whenonly significant predictors of choice (mucositis)are included This reflects the extent of welfare per-ceived from the reduced mucositis and improvedadministration schedule offered by raltitrexed.Most importantly, the information derived fromthe CA can be used to assist both clinicians andgovernment regulators in their appraisal of newtherapies.

Acknowledgements

This study was conducted with the financial assistanceof AstraZeneca, Australia.

References1. Donaldson C. Willingness to pay for publicly provided goods:

a possible measure of benefit. J Health Econ 1990; 9: 103-182. Coley CM, Li Y, Medsger AR, et al. Preferences for home vs

hospital care among low-risk patients with community-ac-quired pneumonia. Arch Intern Med 1996; 156: 1565-71

3. Granberg M, Wiklan M, Nilsson L, et al. Couples willingnessto pay for IVF/ET. Acta Obstet Gynecol Scand 1995; 74:199-202

4. Hudmon SK, Stoltzfus C, Chamberlain RM, et al. Participant’sperceptions of a phase I colon cancer chemoprevention trial.Control Clin Trials 1996; 17: 494-508

5. Morrison GC, Gyldmark M. Appraising the use of contingentvaluation. Health Econ 1993; 1: 233-43

6. O’Brien B, Novosel S, Torrance G, et al. Assessing the eco-nomic value of a new antidepressant: a willingness-to-payapproach. Pharmacoeconomics 1995; 8 (1): 34-45

7. Olsen JA, Donaldson C. Helicopters, hearts and hips: usingwillingness to pay to set priorities for public sector health careprogrammes. Soc Sci Med 1997; 46: 1-12

8. Ritchie LD. A willingness-to-pay analysis: comparison of thepatient preference for topical flurbiprofen local action trans-cutaneous patches and piroxicam gel. Br J Med Econ 1996;10: 291-302

9. Cattin P, Wittink D. Commercial use of conjoint analysis: asurvey. J Marketing 1982; 46: 44-53



10. Farrar S, Ryan M. Response-ordering effects: a methodologicalissue in conjoint analysis. Health Econ 1999; 8: 75-9

11. Ryan M. Economics and the patient’s utility function: an appli-cation to assisted reproductive techniques [dissertation]. Ab-erdeen: University of Aberdeen, 1999

12. Ryan M. Using conjoint analysis to take account of patient pref-erences and go beyond health outcomes: an application to invitro fertilisation. Soc Sci Med 1999; 48: 535-46

13. Ryan M, Hughes J. Using conjoint analysis to value surgicalversus medical management of miscarriage. Health Econom-ics Research Unit Discussion Paper No. 06/95. Aberdeen:University of Aberdeen, 1995

14. Ryan M, McIntosh E, Shackley P. Methodological issues in theapplication of conjoint analysis in health care. Health Econ1998; 7: 373-8

15. Ryan M, Farrar S. Using conjoint analysis to elicit preferencesfor health care. BMJ 2000; 320 (7248): 1530-3

16. Aristides M, Mitchell A, Henry D. Economic evaluation and thesubsidisation of pharmaceuticals. In: Davis P, editor. Con-tested ground: new directions in drug regulatory policy. NewYork (NY): Oxford University Press, 1996

17. O’Brien BJ, Goeree R, Gafni A, et al. Assessing the value of anew pharmaceutical: a feasibility study of contingent valua-tion in managed care. Med Care 1998; 36: 370-84

18. Cunningham D, Zalcberg JR, Rath U, et al. Tomudex(ZD1694): results of a randomised trial in advanced colorectalcancer demonstrate efficacy and reduced mucositis and leu-copenia. Eur J Cancer 1995; 31A: 1945-54

19. Ryan M, Hughes J. Using conjoint analysis to assess women’spreferences for miscarriage management. Health Econ 1995;6: 261-73

20. Greene WH. Econometric analysis. 3rd ed. New York (NY):Prentice-Hall International, 1997

21. White H. A heteroscedasticity-consistent covariance matrix es-timator and a direct test for heteroscedasticity. Econometrica1980; 48: 819-930

22. White H. Maximum likelihood estimation of misspecified mod-els. Econometrica 1982; 50: 1-25

23. Binder DA. On the variances of asymptotically normal estima-tors from complex surveys. Int Stat Rev 1983; 51: 279-92

24. Kish L, Frankel MR. Inference from complex samples. J RoyStat Soc B 1974; 36: 1-37

25. Bryan S, Buxton M, Sheldon R, et al. Magnetic resonance im-aging for the investigation of knee injuries: an investigationof preferences. Health Econ 1998; 7: 595-603

26. Gujarati P. Basic econometrics. 2nd ed. New York (NY):McGraw-Hill Book Company, 1988

27. Magat WA, Viscusi WK, Huber J. Paired comparison and con-tingent valuation approaches to morbidity risk valuation. JEnviron Econ Manag 1988; 15: 395-411

28. Propper C. Contingent valuation of time spent on NHS waitinglists. Econ J 1991; 100: 193-9

29. Cocconi G, Cunningham D, Van Cutsem E, et al. Open, rando-mised, multicentre trial of raltitrexed vs fluorouracil plus highdose leucovorin in patients with advanced colorectal cancer.J Clin Oncol 1998; 16: 2943-52

30. StataCorp. Stata statistical software: release 6.0. College Sta-tion (TX): Stata Corporation, 1999

31. Drummond M, Aristides M. The Australian cost-effectivenessguidelines: an update. In: Towse A, editor. Guidelines for theeconomic evaluation of pharmaceuticals: can the UK learnfrom Australia and Canada? London: Office of Health Eco-nomics, 1998

32. Henry DA. Economic analysis as an aid to subsidisation: thedevelopment of Australian guidelines for pharmaceuticals.Pharmacoeconomics 1992; 1 (1): 54-67

33. Davey P, Grainger D, MacMillan J, et al. Economic evaluationof insulin lispro vs neutral (regular) insulin therapy using awillingness-to-pay approach. Pharmacoeconomics 1998; 13(3): 347-58

Correspondence and offprints: Dr Mike Aristides, M-TAGLimited, Level 3, Lee House, 109 Hammersmith Rd, Ham-mersmith, London, W14 OQH, UK.E-mail: [email protected]



conjoint analysis of a new chemotherapy

Documents