Does the whole equal the sum of the parts? Patient-assigned utility scores for IBS-related health...

HEALTH ECONOMICS

Health Econ. 15: 543–551 (2006)

Published online 3 January 2006 in Wiley InterScience (www.interscience.wiley.com). DOI:10.1002/hec.1074

Does the whole equal the sum of the parts? Patient-assignedutility scores for IBS-related health states and pro¢les

John Braziera, Paul Dolana, Korina Karampelab and Isabel Towersa,*aHealth Economics and Decision Science, School of Health and Related Research, University of Sheffield, UKbGlobal Health Outcomes, GlaxoSmithKline, UK

Summary

The quality-adjusted life year (QALY) assumes that the value of a health state is linearly related to the time spent init, which implies that the value of a health state is independent of the states which precede or follow it. Irritablebowel syndrome (IBS) is a suitable condition to test this assumption since it is subject to considerable fluctuationsover time.Forty-nine IBS patients were asked to rate their own health using generic measures of health and a condition

specific classification. They were then asked to value five IBS states and four profiles using a self-completed versionof the standard gamble technique. The implied value of each profile was estimated using the QALY assumption oflinearity over time and compared with the direct profile valuations.The directly elicited profile values suggest that reductions in the duration of IBS symptoms has less of an impact

on the value of quality of life than would be implied by the QALY assumption of linearity over time, though thedifferences were small. There are a number of competing explanations for this finding, including possible sequenceeffects, quantity effects or time preference, or it might be due to gestalt effects resulting in a neglect of time spent insymptomatic states of health. Copyright # 2005 John Wiley & Sons, Ltd.

Keywords quality-adjusted life years; valuation of health states versus profiles; irritable bowel syndrome

Background

A commonly used measure of benefit in economicevaluations is the number of quality-adjusted lifeyears (QALYs) that an intervention generates. TheQALY combines the value of changes in health-related quality-of-life (HRQoL) with informationon length of life into a single index number. Thedemonstration of cost-effectiveness in terms of anincremental cost-per-QALY ratio is becomingimportant for public funding of new health careinterventions in many countries [1–3]. This paperexamines the application of a QALY approach toirritable bowel syndrome (IBS).

Functional gastro-intestinal disorders accountfor a significant proportion of primary care andhospital outpatient visits. IBS is the most commonof these disorders. Patients diagnosed with IBStypically present with a wide range of symptoms,and there is growing evidence that they experiencea level of HRQoL significantly below that of thegeneral population. An important feature of thiscondition is that its symptoms tend to fluctuateconsiderably over relatively short periods of time.This has important implications both for theassessment of HRQoL in IBS patients and forthe extent to which the QALY can represent thebenefits associated with their treatment.

Copyright # 2005 John Wiley & Sons, Ltd.Received 25 July 2003

Accepted 21 October 2005

*Correspondence to: Health Economics and Decision Science, School of Health and Related Research, University of Sheffield,Regent Court, 30 Regent Street, Sheffield S1 4DA, UK. E-mail: [email protected]

The use of QALYs to value the benefits oftreatment requires a number of quite stringentassumptions about the way people value a profileof health over time [4]. For example, the numberof QALYs associated with a particular healthprofile, consisting of a given health state, Q, anda given number of years, T, is typically calculatedfor use in incremental cost-effectiveness analysisby multiplying the value of Q by T. Thus, thevalue of a health state is assumed to be linearlyrelated to the time spent in that state. This QTQALY model assumes a zero time preference. Forhealth profiles characterised by changes inHRQoL, it is also usually assumed in QALYcalculations that the value attached to a particularhealth state is independent of the state(s) thatprecede or follow it. This is equivalent to theestimation of the area under the curve for repeatedassessments in a clinical trial [5]. There is someevidence that the value implied by summing thevalues for each of the health states within a profiledoes not provide a good approximation of thevalue of the profile when it is measured directly asa whole [6]. If these assumptions are violated, thenthis QALY algorithm may give a false impressionof the value associated with different healthprofiles, and hence the relative effectiveness ofdifferent interventions might then be misrepre-sented [7].

Despite recent interest in the issue of whethersumming QALY scores derived from constituenthealth states of a profile gives an accurate value fora profile, there has been relatively little empiricalresearch in this area for use in economic evalua-tion. Research in the QALY area has focusedlargely on the sequence effects of the constituenthealth states. Richardson and colleagues [6] foundthat profiles, which deteriorate over time, arevalued lower than those which vary little overtime. Lipscomb found that profiles which improvewith time, and finish on a relatively highly valuedstate, are valued higher than those which deterio-rate or show little variation [8]. This suggeststhat the sequences of states has a systematic effecton the value of a profile. However, a study byMackiegan and colleagues showed that, wherethere is very gradual deterioration, sequence hasno significant effect on profile valuations [9].

There have been more investigations of prefer-ences over time in the psychological literature.There is evidence, for example, that people preferprofiles that improve over time and end on a goodnote to profiles with the reverse pattern [10]. It has

also been found that preferences for improvingsequences are moderated by prior expectations ofwhat is realistic [11]. A more general explanationrelevant to the situation of IBS is that profiles maycontain gestalt points, which are given particularweight by the valuer. In particular, the final point,the peak point, and the rate of change inmeasurements over the profile have been foundto be significant predictors of the overall rating ofthe profile [12–14].

The psychological literature has shown thateach part of the profile is not of equal value.Despite this research into the effects of durationand sequences on preferences, there has been verylittle research in the context of health andspecifically the QALY assumptions of linearityover time. There has also not been an investigationof these phenomena in conditions such as IBS thatoffers a particularly interesting example due to thevariable nature of the condition.

The main aim of this study was to comparedirectly elicited values for IBS health profiles withthe values implied by the QALY algorithm for thesame profiles. Should the study find direct andindirect valuations of the scenarios be consistent,then this implies that none of the violationsdescribed above have occurred. However, shouldthey prove to be inconsistent in a statisticallyand decisionally important way, our analyses willbe able to identify the discrepancy but not itsdeterminants.

Methods

The IBS health states and profiles

For this study IBS has been described in terms oftwo symptoms: whether or not they had experi-enced adequate relief of abdominal pain anddiscomfort (P) in the last seven days, and for thenumber of those days on which they felt a sense ofurgency (U). Urgency is described as a dichot-omous variable of more than three days and threedays or less in order to construct health states. Inaddition, there was interest in constipation (C) as aside effect of treatment. These three key symptomsform eight possible health states. Of these, six wereselected for use in this study to keep the number ofvaluation exercises to a minimum to avoidcognitive overload. The two excluded states

Copyright # 2005 John Wiley & Sons, Ltd. Health Econ. 15: 543–551 (2006)

J. Brazier et al.544

contained symptoms of constipation. The sixstates considered in this study are presented inTable 1. Hereafter, the presence of a symptom isindicated by a ‘+’ sign and the absence of thesymptom by a ‘�’ sign; thus P�U�C� is the bestIBS-related health state and PþU þ Cþ theworst state. It is important to bear in mind thatbeing in the best IBS-related health state does notimply the individual is in full health.

The aggregation of health states to form timeprofiles of health was less straightforward. Pre-vious work constructing health profiles has tendedto deal with scenarios where patients are expectedto go through a clear sequence of states. However,IBS does not follow a clear course of progression.Rather, it is characterised by considerable fluctua-tions in HRQoL over relatively short periods. Inaddition, symptoms come and go in patterns thatvary widely across patients. One way of summar-ising the patient’s health experience using the eightIBS health states is in terms of the proportion oftime spent in each state over a set period. Asuccessful treatment may not entirely eliminate thepatient’s experience of a given state, but mayreduce the amount of time spent in that state. Theprofiles are described in terms of the numbers ofweeks in which a patient would be in each IBS

state over a 12 week period, which will be repeatedfor the remainder of the respondent’s life. Thereare four profiles used in this study (Table 2). Theinterview emphasised that it is not known preciselyhow the overall length of time in each state isdistributed over the 12 week period.

Valuation survey

The valuation exercise. The standard gamble (SG)technique has been used to elicit preferences forthe health states and profiles [15]. Basically, the SGasks the respondent to choose between thecertainty of an intermediate health state and theuncertainty of a treatment with two possibleoutcomes, one of which is better than the certainoutcome and one of which is worse. The worststate is usually immediate death, though it doesnot need to be. The health states in this survey aredescribed as lasting for the rest of their life. Theobjective is to find the probability, p, at whichthe respondent is indifferent between acceptingthe intermediate health state and acceptingthe uncertain treatment. The value for the inter-mediate health state is then simply given by p.

Table 1. The six IBS health states

Health-state description Code

You have adequate relief of abdominal pain and discomfort. P�U�C�You do not feel a sense of urgency more than three days per week.

You do not have adequate relief of abdominal pain and discomfort. P+U�C�You do not feel a sense of urgency more than three days per week.

You have adequate relief of abdominal pain and discomfort. P�U+C�You feel a sense of urgency more than three days per week.

You do not have adequate relief of abdominal pain and discomfort. P+U+C�You feel a sense of urgency more than three days per week.

You do not have adequate relief of abdominal pain and discomfort. P+U�C+You do not feel a sense of urgency more than three days per week.You experience constipation.

You do not have adequate relief of abdominal pain and discomfort. P+U+C+You feel a sense of urgency more than three days per week.You experience constipation.

Immediate death

Patient-Assigned Utility Scores for IBS-Related Health States and Pro¢les 545


The version of SG used in the study wasoriginally developed by Jones-Lee and colleagues[16]. This version lists values for the chances ofsuccess from 0 in 100 to 100 in 100 andrespondents are asked to place a tick againstall the probabilities of success at which they areconfident they would choose the treatment and across against all the values where they would rejectthe treatment. The space between the ticks andcrosses indicates the region of indifference. If thisregion covers more than one probability value,then a mid-point is taken to be the indifferencevalue. This version of SG has been found toproduce more consistent and reliable data than aninterview-based variant using props [15].

Many of the IBS health states would beregarded as comparatively mild and so manyrespondents might be unwilling to accept even avery small chance of dying should the treatmentfail. To improve the sensitivity of the SG questionit was decided to change the worst treatmentoutcome in the SG question from death to PþU þ C� for valuing two of the states (i.e. PþU � C� and P�U þ C�) and three of theprofiles (i.e. A, B and C). State PþU þ C� wasin turn valued against the best state and death.This allowed for the two states and three profilesvalued against PþU þ C� to be ‘chained’ ontothe conventional full health-to-death scale used forcalculating QALYs. If the health state valueobtained in the first gamble is U1 and thevalue of the reference state ðPþU þ C�Þfrom the second gamble is U2, the formula forchaining onto the 0–1 scale is simply:U3 ¼ U1 þ ð1�U1ÞðU2Þ.

The sample. The inclusion criteria for this valua-tion study were women aged 18 or over who hadbeen diagnosed with IBS based on the Romecriteria. The aim was to recruit a sample of 50

respondents from four ambulatory care centres inUK. This sample size was based on the ability todetect a difference of about 0.1 (on a 0–1 scale)between different health states for an alpha of 5%and power of 80%. There is no real consensusabout what difference is to be considered mean-ingful but a difference of this kind is likely to beimportant in many decision-making contexts [17].

The interviews. Those eligible for inclusion in thestudy were interviewed in small groups of 2–6 by atrained interviewer (IT). Whilst some studies haveelicited health state valuations using postal ques-tionnaires, there is a great deal of evidence tosuggest that this is an unreliable way to gathersuch data, not least because respondents ofteninterpret questions differently than intended [18].By bringing respondents together in small groups,it was possible to check that they understood theself-completed questionnaire. Respondents in thegroup were asked at the end of a detailedexplanation given by the interviewer whether theyunderstood the task. They were encouragedthroughout to ask for clarification at anytime.The risk of a ‘bandwagon’ effect was minimal sincethe questionnaire was self-administered andthe interviewer was there to answer questions.In practice, there was little dialogue betweenrespondents. Although this process made itpossible for the interviewer to influence thevaluations of respondents, it was decided that thisrisk was worth the gain in respondent under-standing. A pilot study was also conducted toensure that the interview procedure was under-stood as intended.

Respondents were taken through a question-naire, which was to be self-completed. Thequestionnaire began by asking respondents forsome background information relating to theirage, occupation, age when they completed educa-

Table 2. The health profiles

Number of weeks (out of 12) spent in each state in profile

Health states A B C D

P+U+C� 2 4 6 2P+U�C� 2 2 2 2P�U+C� 2 2 2 2P�U�C� 6 4 2 6Constipation } } } 2a

aThese two weeks are distributed over the 12 weeks in ways described in the text.



tion, the length of time they had experienced IBSfor, and a general health question. They were thenasked to indicate their current experience with thethree IBS symptoms. It was made clear to them allthat these questions formed the basis of the healthstates they would be asked to value. At this stage,respondents had the opportunity to ask questionsrelating to how the descriptions should be inter-preted.

Respondents were then asked to rank the statesand profiles in order to familiarise themselves withthe descriptions. They were then taken through thefirst SG question in order to help them understandthe exercise, at which point they were given theopportunity to ask any questions. Once they weresatisfied with what they were being asked to do,they were asked to complete the remainder of thequestionnaire. The questionnaire asked them tovalue five IBS-related states (the two mildestagainst a treatment failure state of PþU þ C�and the remaining three against death) plus fourprofiles. The best state formed the upper anchorand so was not valued.

Respondents were also asked to rate their ownhealth using EQ-5D [19,20]. Once everybody in thegroup had completed the questionnaire, respon-dents were provided with the opportunity tocomment on the exercise.

The analysis of the study data

The first phase of analysis concerns the quality ofthe data. An important check on the extent towhich respondents were able to understand thevaluation tasks is to examine the logical consis-tency of their SG responses. For some pairs of IBShealth states, one state can be regarded as logicallybetter than another if it is better on one or more ofthe symptoms and no worse on any symptom. It ispossible for such pairs of states to have the samevalue, but the logically better state should not begiven a lower score. Similarly, for pairs of timeprofiles, one profile is logically better than anotherwhere it contains less time in one symptomaticstate and no more time in an other symptomaticstate. The number of patients in the sample whodid not meet this definition of consistency wascalculated.

To calculate the implied profile values using theQALY algorithm, each of the constituent healthstate values is multiplied by the overall proportion

of time spent in that state. This is straightforwardfor profiles A–C, which do not contain periods ofconstipation. However, profile D includes twoweeks with constipation but it does not specifywhen this might occur. One solution is to assumethat the impact of each symptom is additive, andthus the state of constipation should be given thesame value regardless of what other symptoms arepresent. The periods of constipation were thereforeassigned to one week of state PþUþ and oneweek of PþU�. However, the impact of con-stipation on health state values may not beadditive.

The statistical analyses included the calculationof the descriptive statistics for the values of IBSstates and profiles, and differences between directand indirect valuations were examined by t-test.

Results

In total, 49 respondents took part in this study.The interviews took between 30min and 1 h tocomplete, with larger groups taking longer. Themean age of the sample was 47 and the averagenumber of years with IBS was 25. Of the women,74% were in paid employment. In terms of the IBSsymptoms, 34% reported pain and discomfort,72% felt a sense of urgency for more than threedays in the past seven days, and 36% experiencedconstipation. The mean EQ-5D generic healthstatus score for the sample using Tariff 1 from theYork MVH survey of the UK general populationwas 0.66 [20].

The logical consistency conditions in the healthstate valuation component of the study arePþU þ C� > PþU þ Cþ and P+U�C+>P+U+C+. There are very few violations ofthe consistency conditions relating to these healthstates, with rates of 10 and 8%, respectively, andthese are comparable with other studies [11]. It isnoteworthy that there are more violations ofconsistency in the direct profile values. The logicalconsistency conditions are A>B, A>C, B>Cand A>D, with 18, 23, 6 and 25% violations,respectively. However, many of the inconsistencieswere such that the value of the logically worse stateor profile was within 0.05 of the value of thelogically better state or profile at the individuallevel. In addition, the inconsistencies were un-related to respondent characteristics and norespondent was inconsistent throughout.



Table 3 shows the mean values of the five healthstates, where one is the best state defined byabsence of IBS symptoms (i.e. state P�U � C�).Since one respondent had missing data for healthstate PþU þ C�, and some values had to bechained to death through this state, it is sometimesonly possible to produce results for 48 respon-dents. It can be seen that the overall values followa logical ordering and that all mean values aresignificantly below the value for full health,indicating that the presence of IBS symptomshad a significant effect on quality of life.

Table 4 compares the directly elicited profilevalues with the implied values associated withthose same profiles as calculated using the QALYassumption of linearity over time. The directlyelicited profile values were found to be less thanthe implied value for profiles A and D, and more

than the implied value for B and C. For profile Athe difference is nearly significant (p ¼ 0:07), butnowhere near so for profile B. Differences inprofiles C and D were significant at the 5% level.

For the three profiles not containing constipa-tion (i.e. A, B and C) these results show a reversalof the relationship between direct and impliedprofile valuations, with the direct valuation beingless than the indirect valuation for profile A andmore for profile C and nearly the same for profileB. This reversal results from the fact that theimplied and direct valuations of the profiles followa different pattern. Direct profile mean and medianvalues are essentially the same across these threeprofiles (with the value of the health decrementsfrom profile A being �0.002 and 0.000 for profilesB and C, respectively, which were non-significantat the 5% level), whilst the implied values decline

Table 3. Health state values

Health state N Median Mean (SD) Mean difference from 1.0

P�U+C� 49 0.997 0.982 0.018n

(0.988–1.000) (0.04)P+U�C� 49 0.997 0.986 0.014n

(0.990–1.000) (0.03)P+U+C� 48 0.995 0.952 0.048n

(0.980–1.000) (0.10)P+U�C+ 49 0.995 0.952 0.049n

(0.975–1.000) (0.12)P+U+C+ 49 0.985 0.933 0.067n

(0.965–1.000) (0.17)

nDifference is significantly different from 1.000 at p50:05.

Table 4. Direct and implied profile values

Median (IQR) Mean (SD) 95% confidenceinterval around

Paired t-test(2-tail significance)

Profile N Direct Implied Direct Implied the difference p

A 48 0.997 0.998 0.982 0.987 �0.001 to 0.011 0.076(0.992�0.999) (0.992�0.999) (0.05) (0.03)

B 48 0.997 0.997 0.980 0.979 �0.002 to 0.004 0.513(0.992�0.999) (0.987�0.998) (0.05) (0.045)

C 48 0.997 0.996 0.982 0.970 0.005 to 0.019 0.002(0.991–0.999) (0.983–0.997) (0.04) (0.06)

D 48 0.985 0.997 0.937 0.982 �0.083 to �0.007 0.021(0.975–0.995) (0.990–0.998) (0.16) (0.03)



in a linear fashion as would be expected given theQT QALY assumption (with the value of thehealth decrements from profile A of profiles B andC being 0.008 and 0.017 (p50:05)).

Given the comparatively high levels of respon-dent inconsistency, the same analysis of meanvalues was undertaken excluding the 15 respon-dents who were inconsistent in any of the six waysidentified earlier. This did not alter the pattern ofthe relationship between the direct and impliedvaluations.

Profile D includes the additional symptom ofconstipation, and therefore does not enter into theabove comparison. It has the same amount of timein pain and urgency as profile A, but has twounspecified weeks associated with constipation,which run along side the other symptoms andcould occur at any point during the 12 weeks. Themean profile value for profile D is significantly lessthan the mean value of profile A. The directvaluation was also found to be significantly lowerthan the implied valuation.

Discussion

This paper has an important contribution to maketo the literature on the valuation of health for usein economic evaluation. It has been claimed in thepast that the QT QALY algorithm does notadequately represent people’s preference over timeprofiles of health [7]. The study reported in thispaper would appear at first sight to provideevidence of a violation of the linearity assumptionof the QALY model.

Despite the higher levels of inconsistency anddispersion associated with direct valuations it ishard to dismiss this finding as simply an artefact.The result was robust to the exclusion of incon-sistent respondents and was found to be commonacross most respondents at the individual level.The general health status score of the sampleindicated states of health significantly below thatexpected in people of their age [21]. Furthermore,many of them were experiencing the symptomsthey were being asked to value. However, theremay be a concern that the sample of IBS suffererswere involved in an ongoing clinical trial and maynot be perfectly representative of IBS sufferers inthe community at large. This was mainly amethodological study and hence the precise sampleof respondents used is not important.

The SG task did not have death as one of thereference states for five out of the nine valuations.This was done for two reasons. The first was thatthe valuation task needed to be realistic for therespondents to take them seriously. There was alsoa concern that respondents might become alarmedat a reference to death in every SG question.Therefore, for the milder states and profiles, thereference states of P�U � C� and PþU þ C�have been used, where the latter was worse thanthe states being valued. The second reason was toincrease the sensitivity of the scale. However, thispaper is addressing issues around the derivation ofQALYs, where zero represents states regarded asequivalent to death and one states in full health,and it has been necessary to chain the values usingthe valuation of PþU þ C�. This can becriticised due to inconsistency between chainedvalues and direct valuations [22,23]. However, theanalysis has been repeated for profiles A, B and Cusing unchained values and the same relationshiphas been found between direct and implied profilevalues.

It is important for possible users of the resultspresented in this paper to appreciate that thesymptom-free state P�U � C� does not neces-sarily correspond to full health. Whilst we wereable to chain the health state and profile valuationsto death, the upper end of the scale could not beanchored to the end point of full health. This doesnot matter for the methodological comparisonpresented in this paper, but the results could not beused in cost-utility analysis without further adjust-ment [24].

The fact that P�U � C� is not full health mayalso partly explain the discrepancy between meanIBS health state values of 0.933–0.986 and therespondents’ own EQ-5D scores of 0.66. The IBSstates take no account of other aspects of thecondition or co-morbidity. Furthermore, the SGvaluations of the IBS states will be subject to riskaversion. Finally, the EQ-5D scores have beenobtained from a general population sample andthese have been shown to be lower than those ofpatients in many conditions [25].

This paper offers an important insight into theway people value conditions such as IBS, wherethe condition does not follow a clear progression.Previous work in the economics literature onvaluing time profiles of health has tended to focuson simple sequences of states. IBS is characterisedby two distinct features, which have not beenexamined in previous work on valuing profiles.



The first is that IBS is subject to considerable andunpredictable fluctuations in its symptoms andhence HRQoL varies over relatively short periods.Yet at the same time it is a condition that can lastfor many years. We developed a method forsummarising the patient’s health experience usingfour IBS health states in terms of the proportion oftime spent in each state over a set period (12weeks), which will be repeated for the remainder ofthe respondent’s life. The time spent in each healthstate is presented as being scattered in anunpredictable way over each period of 12 weeks.This method allows the outcomes of treatment tobe compared in terms of the average amount oftime spent in each state. This study showed howsuch profiles can be valued directly using SG. IBSis not the only condition with these features andthe results presented in this paper have implica-tions for other medical conditions.

This study adds to the currently limited amountof empirical work in the economic literatureaddressing an important debate between the useof health state values via QALYs and thealternative of the HYE or profile valuation. Theresults provide evidence of the incongruencebetween the QALY assumption and the waypeople directly value time profiles of health. Whilstthe inconsistency was significant in three out of thefour profiles examined, the differences were rathersmall at 0.005 (A), 0.001 (B), 0.012 (C) and 0.045(D). It could be argued that such small differenceswould have little decisional impact. However,these differences need to be measured against avery narrow range of values for these profiles,which range from 0.937 to 0.987. For a patientchoosing between treatments, these differencesmay still prove to be important. The economicimportance of these differences really depends oncontext and while the differences may be small inabsolute terms, they may nonetheless prove to beimportant where the cost differences are also small.

However, this study was not designed to explainthe differences found. The QALY model used inthis study to value the profiles (i.e. the ‘implied’profile values) essentially assumes, following onfrom the distinctions made by Gafni and Tor-rance, no sequence effects, no quantity effects anda zero time preference [26]. The results found inthis study could be partly explained by anyone ofthese. The sequence of states in the profile was notspecified for the direct valuation. This was left tothe respondents to think of possible sequences.The sequence of states would also have implica-

tions for time preference. However, given that weknow neither the sequence imagined by therespondents nor their time preference rate, we donot know the likely direction or size of theseeffects.

It could be a result of a quantity effect wherebyindividual valuations of states is not proportionalto the time spent in them. There could be somekind of threshold effect causing a non-linearpattern in direct valuations. This hypothesis canonly be tested through further research valuingIBS profiles with 8, 10 and 12 weeks in statePþU þ C�.

There are other explanations found outside ofthe health economics literature. One possibleexplanation might be found in the work of Arielyand Zauberman who addressed issues aroundvaluing a whole profile versus segments [27]. Theyshowed that the overall rating of an entire profileof non-health-related events such as annoyingsounds is based on the gestalt characteristics; butwhen the profile is broken into segments which arerated separately, the mean value of each segmentbecomes more important in the overall valuationof the profile. The work by Ariely and Zaubermancould go some way towards explaining thedifference between implied and direct methods ofvaluing IBS health profiles. The implied values areobtained by segmenting the profiles into theirconstituent health states. However, the directvaluations are based on gestalt characteristics,which for the IBS profile could simply be the worsthealth state, rather than the proportion of timethey could expect to be in the state. For valuationsby the implied method a good segment would bevalued highly, and a bad segmented valued low.This would suggest that the implied method wouldvalue a profile that tended to be good more highlyand a profile with a tendency to be bad morelowly. This may account to some extent for thesteps in the valuation of profiles A–C found in thisstudy.

It has been argued by some commentators thatthe direct valuation of profiles is superior to theQALY algorithm since it does not assume people’spreferences are linear over time. This position isbased on conventional choice theory and takes noaccount of the cognitive difficulty associated withthe valuation of profile compared to health states.The greater difficulty associated with valuingprofiles is reflected in the higher inconsistencyrates. The psychology literature suggests thatrespondents in this circumstance may use simplify-



ing heuristic or gestalt devices to overcome thecognitive load, such as concentrating on the state,rather than considering the time spent in the state.For the purpose of valuing benefits of health care,it will be necessary to address these problems withthe direct valuation of profile. Establishing whichmethod offers a better way of valuing healthprofiles ultimately requires some test of externalvalidity, but this requires further research.

Acknowledgements

We are grateful to Glaxo-Smith-Kline for funding thisstudy. We would like to thank the reviewers for theirconstructive comments. The usual disclaimers apply.

References1. Ministry of Health (Ontario). Ontario Guidelines for

the Economic Evaluation of Pharmaceutical Pro-ducts. Ministry of Health: Toronto, 1994.

2. Commonwealth Department of Health, Housingand Community Service. Guidelines for the Pharma-ceuticals Industry on the Submission to the Pharma-ceutical Benefits Advisory Committee. AustralianGovernment Publishing Service: Canberra, 1992.

3. Department of Health and the Association of thePharmaceutical Industry. Guidelines for the Eco-nomic Evaluation of Pharmaceuticals [Press release].Department of Health: London, 1994.

4. Bleichrodt H, Wakker P, Johannesson M.Characterizing QALYs by risk neutrality. J RiskUncertainty 1997; 15: 107–114.

5. Matthews JNS, Altman DG, Campbell MJ, Roy-ston P. Analysis of serial measurements in medicalresearch. Br Med J 1990; 300: 230–235.

6. Richardson J, Hall J, Salkfeld G. The measurementof utility in multiphase health states. Int J TechnolAssess Health Care 1996; 12: 151–162.

7. Mehrez A, Gafni A. Quality-adjusted life years,utility theory, and healthy-years equivalents. MedDecis Making 1989; 9: 142–149; 10(2): 148–149.

8. Lipscomb J. The preference for health in cost-effectiveness analysis.Med Care 1989; 27: S233–S253.

9. Mackeigan LD, O’Brien BJ, Oh PI. Holistic versuscomposite preferences for lifetime treatment se-quences for type 2 diabetes. Med Decis Making1999; 19: 113–121.

10. Varey C, Kahneman D. Experiences extendedacross time: evaluation of moments and episodes.J Behav Decis Making 1992; 5: 169–185.

11. Chapman GB. Preferences for improving anddeclining sequences of health outcomes. J BehavDecis Making 2000; 13: 203–218.

12. Ariely D, Carmon Z. Gestalt characteristics ofexperiences: the defining features of summarizedevents. J Behav Decis Making 2000; 13: 191–201.

13. Kahneman D, Fredrickson BL, Schreiber CA,Redelmeier DA. When more pain is preferred toless: adding a better end. Psychol Sci 1993; 4:401–405.

14. Redelmeier DA, Kahneman D. On the discrepancybetween experiences and memories: patients’ real-time and retrospective evaluations of the pain ofcolonoscopy. Working Paper, 1993.

15. Dolan P, Gudex C, Kind P, Williams A. Valuinghealth states: a comparison of methods. J HealthEcon 1996; 15: 209–231.

16. Jones-Lee M, Loomes G, O’Reilly D, Phillips P.The value of preventing non-fatal road injuries:findings of a willingness-to-pay national samplesurvey. Transport Research Laboratory, 1993.

17. O’Brien BJ, Drummond MF. Statistical versusquantitative significance in the socioeconomic eva-luation of medicines. PharmacoEconomics 1994; 5:389–398.

18. Slovic P. The construction of preferences. AmPsychol 1995; 50(5): 364–371.

19. Brooks R. EuroQol: the current state of play.Health Policy 1996; 37: 53–72.

20. Dolan P. Modelling valuations for EuroQol healthstates. Med Care 1997; 35: 351–363.

21. Kind P. The EuroQol instrument: an index ofhealth-related quality of life. In Quality of Life andPharmacoEconomics in Clinical Trials (2nd edn),Spilker B (ed.). Lippincott-Rivera: Philadelphia,PA, 1996; 191–201.

22. Rutten-Van-Molken M. Costs and effects of phar-macotherapy in asthma and COPD. PhD Thesis,Universitaire Press Maastricht Datawyse Maas-tricht, 1994.

23. Llewellyn-Thomas H, Sutherland HJ, Tibshirani R,Ciampi A, Till JE, Boyd NF. Describing healthstates: methodological issues in obtaining values forhealth states. Med Care 1984; 22: 543–552.

24. Drummond MF, O’Brien B, Stoddart GL, TorranceGW. Methods for the Economic Evaluation of HealthCare Programmes (2nd edn). Oxford MedicalPublications: Oxford, 1997.

25. Brazier JE, Akehurst R, Brennan A et al. Shouldpatients share a greater role in valuing health states?Health Economics and Decision Science DiscussionPaper Series no. 04/3, ScHARR, University ofSheffield, 2004.

26. Gafni A, Torrance GW. Risk attitude and timepreference in health. Manage Sci 1984; 30: 440–451.

27. Ariely D, Zauberman G. On the making of anexperience: the effects of breaking and combiningexperiences on their overall evaluation. J BehavDecis Making 2000; 13: 219–232.



Does the whole equal the sum of the parts? Patient-assigned utility scores for IBS-related health...

Documents

Transcript of Does the whole equal the sum of the parts? Patient-assigned utility scores for IBS-related health...