In r Out..Income Losses

Embed Size (px)

Citation preview

  • 7/29/2019 In r Out..Income Losses

    1/8

    In or Out? Income Losses in Health State Valuations: A Review

    Carl Tilling, MSc,1 Marieke Krol, MSc,2,3 Aki Tsuchiya, PhD,1,4 John Brazier, PhD,1 Werner Brouwer, PhD2,3

    1School of Health and Related Research, University of Sheffield, Sheffield, UK; 2Institute for Medical Technology Assessment, ErasmusUniversity/Erasmus MC, Rotterdam,The Netherlands; 3Department of Health Policy and Management, Erasmus University/Erasmus MC,Rotterdam,The Netherlands; 4Department of Economics, University of Sheffield, Sheffield, UK

    A B S T R A C T

    Background: In 1996 the Washington Panel controversially recommended

    valuing productivity costs (PC) in terms of quality-adjusted life years. The

    Panels assumption that respondents in health state valuation (HSV) exer-

    cises take income losses into account could not be countered since there

    was no evidence regarding what people consider in HSV exercises. If they

    do consider income losses and if this changes HSVs, then all economic

    evaluations that have included PC in the numerator may have double-

    counted these costs. Alternatively, if respondents do not consider income

    losses then all past economic evaluations that have not included PC in the

    numerator have failed to account for sizeable societal costs.

    Objectives: Through a review we aim to recapture the debate surroundingthe appropriate method for including PC in health economic evaluations,

    to identify empirical evidence addressing the assumptions of the Panel, and

    recommend a future research agenda.

    Methods: Through a review we identify, outline, and critically appraise

    the existing empirical studies that attempt to address whether respondents

    include income effects in HSV exercises.

    Results and conclusion: Seven empirical studies were identified. Overall, it

    seems that not explicitly mentioning the inclusion of income will induce a

    minority of respondents to include these effects and this appears not to

    influence results. More empirical work is needed, using generic instru-

    ments, larger samples, and using the interview method of administration.

    Keywords: health state valuation, income effects, productivity costs,quality-adjusted life years.

    Introductionvhe_614 298..305

    Economic evaluations increasingly influence decisions on theallocation of scarce resources within the health-care sector. Thegeneral idea behind these economic evaluations is that health-care programs should offer sufficient value for money in order towarrant their funding. In order to properly determine the valuefor money of health-care programs, from the societal point ofview, it is pivotal that all relevant costs and effects are included inthe evaluation in an appropriate way.

    One area of particular debate is that of productivity costs,which have been defined as costs associated with productionloss and replacement costs due to illness, disability and death ofproductive persons, both paid and unpaid [1]. In a Cost-effectiveness Analysis, changes in productivity costs were typi-cally valued in monetary terms in the numerator of the cost-effectiveness ratio (C/E) through either the human capital (HC)or friction cost (FC) method. However, in 1996, the WashingtonPanel argued that productivity costs are, and indeed should be,included in the denominator of the C/E ratio [2]. They areincluded as health effects according to the Panel, because inhealth state valuation (HSV) exercises respondents are assumedto consider the effect that a health state will have on income (andthese income losses act as a proxy for productivity costs) evenwhen the valuation method is silent regarding income. To includeproductivity costs in the numerator of the C/E ratio in monetaryterms, as commonly done, would result in double-counting thesecosts, which should obviously be avoided. Alternatively, ifrespondents do not (consistently) consider income effects inHSVs or if the influence of such considerations on the valuationsis negligible, then economic evaluations using the WashingtonPanel Approach have excluded real and sizeable societal

    costs. Therefore, to ensure a good distinction between costsand effects in health economic evaluations, in order to avoiddouble-counting and ensure the inclusion of real societal costs,it is pivotal to determine what respondents consider in HSVexercises.

    By now, empirical evidence on this issue has become avail-able. Studies on whether respondents consider income in HSVswhen these are silent on this matter have been published. Some of

    these studies also consider the influence of providing respondentswith explicit information on income losses associated with agiven health state or with explicit instructions on the inclusion orexclusion of possible income changes. These studies thus alldirectly address the issue of inclusion of income losses in HSVs,which is central in the recommendations made by the Washing-ton Panel.

    In this review article we do not add to the existing empiricalwork. We first outline the different approaches to valuing pro-ductivity costs in economic evaluations. We then present thefindings of existing empirical studies that have attempted todetermine whether or not respondents in HSV exercises considerincome effects. In doing so, we will consider the implications ofthe results for the inclusion of productivity costs and the elicita-tion of HSVs in economic evaluation. Moreover, we will expose

    unresolved issues and hence research priorities for the future. Incontrast to previous studies [35], this article covers the recentempirical developments on the inclusion of income changes inHSVs.

    Background

    The inclusion of productivity costs in economic evaluations hasbeen and still is controversial. When a broad societal perspec-tive is adopted, as is often recommended [2,6], and whichfollows logically from the welfare theoretical roots of economicevaluations [7], all costs and effects should be considered in ananalysis. In this article we do not address the normative question

    Address correspondence to: Carl Tilling, School of Health and Related

    Research, Regent Court, 30 Regent Street, Sheffield, S1 4DA, UK. E-mail:

    [email protected]

    10.1111/j.1524-4733.2009.00614.x

    Volume 13 Number 2 2010

    V A L U E I N H E A L T H

    298 2009, International Society for Pharmacoeconomics and Outcomes Research (ISPOR) 1098-3015/10/298 298305

  • 7/29/2019 In r Out..Income Losses

    2/8

    of whether productivity costs should be included in economicevaluations. Rather, we simply address the issue of how toinclude them in an appropriate way, focusing on the question ofwhether productivity costs (or rather income changes) are, orcan, be included in HSVs.

    Health economic evaluations normally take the specific formof cost-effectiveness analyses or, preferably, cost-utility analyses.

    The latter type of analysis uses the Quality-adjusted Life Year, orQALY gained [8] as a measure of health improvements. Impor-tantly, these analyses, unlike traditional cost-benefit analyses,require the determination of what is counted as a health effect(nonmonetarily) and what as costs. Moreover, as in any analysis,double-counting of impacts needs to be avoided and an adequatemeasurement and valuation method needs to be applied. In thatcontext, it was (and largely still is) common practice and consid-ered most appropriate to incorporate productivity costs in theanalysis as costs in the numerator of the cost-utility ratio.However, in 1996 the Washington Panel controversially recom-mended the inclusion of productivity costs almost entirelythrough HSVs and therefore in terms of quality of life (QoL) inthe denominator of the C/E ratio [2].

    Valuing Productivity Costs in Monetary TermsTheHuman Capital (HC) and Friction Cost (FC) MethodsTwo main methods for valuing productivity costs existed beforethe US Panel proposed their third way. The first is the HCapproach [9,10]. Under this approach lost production (oftenrelated to paid work) as a result of morbidity or mortality isvalued by measuring time lost from work and multiplying thiswith the gross wage of the involved individual. Economic theorysuggests that under certain conditions, at the margin, grosswages equal the productive value of individuals, so that thismultiplication should yield a good estimate of the value of lostproduction. The relevant period of time over which costs (orsavings) are measured is, unless restricted by the time horizon ofthe analysis, the total period of time in which a person is

    (un)able to be productive compared to the alternative scenario.In the case of disability or premature death this can obviouslyamount to many years i.e., until the retirement age would havebeen reached. If a 45-year-old individual earning 30,000 peryear became unable to work due to permanent disability, theestimated value of productivity losses under the HC method(assuming a retirement age of 65) would be: 20 30,000, i.e.,600,000. For examples of economic evaluations that have usedthe HC approach, see Beutels et al. [11], Ford et al. [12], andKobelt et al. [13].

    The neoclassical theory on which the HC approach is basedassumes that all markets clear. However, in reality the existenceof involuntary unemployment means that when someone isforced to leave the workforce due to illness they can be replacedby a previously unemployed person. To be able to explicitly value

    productivity costs from a societal perspective and in relation toeconomic circumstances, the FC method was developed [1416].Under this approach the period in which productivity costs occuris limited to the time it takes to replace a worker. The FC methodargues that, from a societal point of view, there are no productionlosses in the long run, since the production loss in the ill, dis-abled, or deceased worker is cancelled out by a production gainin the new, formerly unemployed, worker. Estimates of the valueof productivity change according to the FC method do includesome additional costs such as the resource cost associated withrecruiting and training replacement workers (e.g., advertising thejob vacancy). Unsurprisingly, the estimates of the FC and HCmethods do not differ significantly for short-term absence [17].

    However, in the case of mortality and long term morbidity thesedifferences are substantial. The estimated value of productivitylosses for society in the case of an individual permanently dis-abled, mentioned above, would be considerably lower for the FCmethod than for the HC method. If it took 1 year to find areplacement worker, and there were advertising and trainingcosts of 10,000, the total cost would be 40,000 (vs. 600,000

    in the HC approach). For examples of economic evaluations thathave used the FC approach, see Glick et al. [18], Van Enckevortet al. [19], and Nord et al. [20]. The criticism of the FC methodmainly relates to the valuation of leisure time; see Johannessonand Karlson [21], Liljas [22], Koopmanschap et al. [23], Johan-nesson [24], and Brouwer et al. [25].

    Valuing Productivity Costs as Part of the Health Effects:The Washington Panel ApproachThe Panel asserts that in valuing imperfect health states, respon-dents will consider their productivity and income level. In otherwords, the Panel assumes that when answering HSV questions(for example, the Time Trade-off [TTO] [26]) respondents willtake account of the effect a potential health state will have not

    only on their ability to work as a form of role functioning [25]but also on their income. The Panel thus proposes that produc-tivity costs can, should, and, in fact, already are included inutility scores for imperfect health states. Most HSV instrumentsare silent on the topic of income changes (with the exception ofthe Health Utilities Index [HUI] instruments), but the Panelargues this is sufficient for inclusion of these effects. It is worthnoting that the Panel does not deny the possibility of replacementof ill workers. It recommends replacement costs be calculatedand included in the numerator. For an example of an economicevaluation that has used the Washington Panel Approach, seeWang et al. [27].

    If we assume that respondents in health states valuations onaverage lower their valuation of some health state by 0.05 (on ascale with 1 for full health and 0 for dead) and the permanently

    disabled individual lives for another 20 years to become 65, thenone additional QALY is being lost (i.e., 20 0.05). A treatmentthat will enable him to return to work will thus gain one addi-tional QALY relative to QALYs gained without income beingconsidered.

    The Washington Panel Approach has received considerablecriticism. An individuals income may be a poor proxy for hisproductivity and hence the stable relationship implicitly assumedby the Panel may not exist [1,25,28]. Productivity changes maynot cause income changes proportional to productivity changes ifsocial benefits (or payments from private insurance) are receivedto compensate for the reduction in labor income. Thus, theimpact of productivity on income may fluctuate with age (e.g.,after retirement, income is likely to be independent of produc-tivity). It may also be difficult for respondents to accurately

    predict how (generic) health states affect productivity and, con-sequently, income. Moreover, illness may cause productivitylosses without absence (i.e., reduced productivity at work orpresenteeism). Brouwer et al. [1] argued that adoption of theWashington Panel Approach would lead to an omission of thesecosts since these productivity losses often do not translate intoincome losses. Furthermore, Meltzer and Johannesson [29] notethat if people (are to) incorporate personal financial conse-quences such as income losses into QALY weights, they willpresumably also incorporate other consequences such as out-of-pocket medical expenses related to, for example, pharmaceuti-cals or hospitalization (whenever relevant) and the value of timeforgone participating in health interventions. However, since the

    In or Out? 299

  • 7/29/2019 In r Out..Income Losses

    3/8

    Panel advises including both of these factors in measures of costs,this would then result in double-counting. Further criticism ofthe Washington Panel Approach, such as the failure to takeaccount of the income gain achieved by the previously unem-ployed member of society, can be found in Brouwer et al. [1]. Ingeneral, it seems likely that respondents in HSV exercises take anindividual rather than a societal perspective, so that costs and

    gains in others (employers, replacement workers, etc) will remainunvalued.

    In or Out? Empirical Studies So Far

    To our knowledge seven empirical studies on this issue have beenpublished so far [3036]. To confirm there were no other studiesthat we were unaware of we performed a Medline search (viaPubmed). We used a title and abstract search with the followingterms: health state valuation OR TTO OR SG OR VAS ORquality of life AND (lost income OR income effects OR incomeloss OR productivity costs) AND consider OR included ORinclusion OR instruction OR double-counting. This gave 297results. The inclusion criteria were any study that had empiricallytested the inclusion of income in HSVs. All but one of the studies

    that we were aware of appeared in the search. The one that didnot appear [30] is only available as an abstract, and hence doesnot appear in Medline. We did not find any other empiricalstudies. Table 1 provides an overview of the key findings andcharacteristics of the seven empirical studies.

    The earliest study was performed in the United States byMeltzer and colleagues [30], asking 402 subjects to value backpain and 429 to value blindness using the TTO method.Respondents were randomly assigned one of three different ver-sions of TTO: in version 0, no guidance was given about finan-cial consequences; in version 1, respondents were told that 60%of their current income would be provided as disability pay-ments; and in version 2, they were told that there would be nodisability payments. When questioned after the TTO, of respon-dents presented with version 0 of the questionnaire, less than

    15% indicated they had considered the financial consequencesof either health state valued. This means that what has beenlabeled as spontaneous inclusion of income when the TTOquestion is silent on the issue is relatively uncommon, unlike thesuggestion of the Panel. Even for version 2, this figure was stillless than 25%.

    One would expect version 2 to elicit the lowest value, whilemaking predictions for versions 0 and 1 is more difficult.However, given that so few people include income effects, even inversion 2, any differences between the versions are likely to besmall. The mean TTO scores for blindness were 4.95, 4.83, and4.84 for versions 0, 1, and 2 respectively (insignificant). Themean TTO scores for back pain were 5.09, 5.78, and 4.25 forversions 0, 1, and 2, respectively. When comparing with version0, the results for back pain were significantly different at the 10%

    and 5% significance levels for versions 1 and 2, respectively.Unfortunately the differences in results between those that didand did not consider income effects are not presented. The sig-nificant results for back pain confirm that version 2 elicits thelowest utility value. However, the relatively high value forversion 1 (back pain) is somewhat surprising, which demon-strates the need for further exploration of why subjectsresponded to the HSV exercises in the way that they did.

    The main conclusion of the authors is that the economiccosts of illness are unlikely to be reflected in QoL questions, sothese need to be counted separately (p. 517). Furthermore theyargue that it may be best to instruct people to ignore theeconomic consequences of illness in answering QoL questions

    (p. 517). Explicit instructions were proposed as a means ofbreaking the silence in order to ensure consistency betweenrespondents regarding what they consider in conventional HSVs.This study was only published as an abstract, in which thespecific survey method is not mentioned, the exact scale of theTTO values is not explained, and no background characteristicsof the sample are given (age and occupation would be particu-

    larly useful in interpreting income effects).A second study was performed by Sendi and Brouwer [31],who presented 20 health professionals (5 medical doctors, 2medical researchers, and 13 nurses), in Switzerland, with a ques-tionnaire that described the health status of a 30-year-old malepatient suffering from multiple sclerosis. They were asked to ratethe health state on a visual analogue scale (VAS) ranging from 0(worst possible health) to 100 (best possible health). The ques-tionnaire was silent on the conditions effect on income (as wellas on leisure). After answering the VAS question (the ex antevaluation), respondents were asked whether the impact illhealth may have on income was included in their valuation. Ifrespondents gave a negative response, they were explicitly askedto consider income (and leisure) effects in the valuation ofthe health state in an identical second VAS question (ex post

    valuation), but the actual magnitude of income loss was notspecified.The results show that 60% of respondents did not consider

    the effects of ill-health on income in the first instance (i.e., whenthe measure was silent), and, as expected, those who did notconsider income had a significantly higher mean VAS score com-pared with those who did consider income (48.33 vs. 31.25),although these groups also varied in terms of the inclusion ofleisure. The ex post valuation for those respondents who did notinclude leisure and income in the first round was significantlylower than their ex ante valuation, but this effect seems especiallyto be driven by leisure considerations. As acknowledged by theauthors, this study has a small sample size weakening any poten-tial conclusions. Also, in this study, background characteristics ofthe sample are not provided. The study moreover used a VAS as

    the valuation component. Since the aim was to find out whetherpeople would include income effects in conventional valuationexercises, the VAS was anchored to best and worst imaginablehealth, meaning that, conceptually, only if respondents considerincome to be a part of health would they include it in valuationson this scale.

    A third study was that by Krol and colleagues [32]. Theyadministered HSV questionnaires, using a similar VAS scale asSendi and Brouwer [31] did, to 227 members of the generalpopulation in The Netherlands. Besides looking at spontaneousinclusion of income (and leisure), this study also addressed thetopic of the effects of explicit prior instructions regarding inclu-sion of income effects on HSVs. The aim of the study thereforewas to test: 1) whether or not respondents spontaneously includethe effect of ill health on income (and leisure); 2) the impact on

    the valuation of inclusion (or exclusion) of income effects; and 3)the influence of explicit instruction on this matter. To test this,three versions of the questionnaire were administered. The firstversion included no directions and those who spontaneouslyincluded income effects were asked to revalue the state again butexcluding these effects. (Note that this is exactly opposite to theex post valuation in Sendi and Brouwer [31], who asked peoplewho did not include income in the ex ante phase, to revalue thehealth states now includingincome. Moreover, the ex post valu-ation now pertained to income only, not leisure). In the secondversion respondents were explicitly instructed up front to incor-porate income effects in their valuations, while in the thirdversion respondents on the contrary were explicitly instructed to

    300 Tilling et al.

  • 7/29/2019 In r Out..Income Losses

    4/8

    Table1

    Summaryofempiricalstudies

    Study

    Methodof

    preference

    elicitation

    Healthstatesvalued

    Sample

    Size

    Sampledemo

    graphic

    Location

    Spontaneously

    includedincome

    effects

    Meanscore

    spontaneously

    considering/not

    consideringincome

    Meanscore

    informed/uninformed

    onincome-effects

    Meanscoreexplicitly

    instructedto

    include/exclude

    income

    Meltzer,Weckerle,and

    Chang(1999)[30]

    TTO

    Blindnessandbackpain

    831

    Patients

    UnitedStates

    $1900 per month). As the authors acknowledge, such a sampleis not representative of the general public, which may lead tobiased results. For instance, one might expect more educatedrespondents to give more thought to the valuation exercise andfully consider the potential consequences of a given state, andhence be more likely to consider income effects. On the otherhand, they might also think more about the concept of health andpurposely leave out income effects from the valuation on a VASmeasuring health. In the case of higher income respondents, thesemay be more likely to exhibit noticeable income effects, sincethey stand to lose more money in absolute terms due to ill health,but on the other hand, they may normally have jobs in which atleast some forms of health problems will less quickly influencetheir productivity. Therefore, making clear predictions about theinfluence of these background characteristics is quite difficult.

    For the experiment of comparing the three versions of thequestionnaire, the biased sample should have had little effect onthe results since the groups that did and did not consider income(both instructed and spontaneously) did not differ in this respect.In the absence of instruction 36% of respondents includedincome effects (comparable to the results of Sendi and Brouwer[31] above), but valuations of the two groups were not signifi-cantly different (contradicting the findings of Meltzer and col-leagues [30] and Sendi and Brouwer [31]). When those thatindicated that they had considered income effects in version 1were asked to repeat the exercise assuming no income effects, nosignificant differences in valuations were observed in the case ofstate 1. This is not surprising, given that the majority of respon-dents did not think this state would reduce their income. In thecase of states 2 and 3, valuations did change significantly (revised

    upwards). However, over half of the respondents did not changetheir valuations (which were also observed for income effects inthe study of Sendi and Brouwer [31]) and this held for all threestates. Across all respondents there were no significant differ-ences between valuations using version 2 or version 3, suggestingthat explicit instruction does not matter in the sense that it doesnot change the overall results, and this again holds for all threehealth states. Therefore, instructing respondents to either includeor exclude income effects in their HSVs in this study did notresult in a noticeable difference between the groups. This castsserious doubts on whether the Washington Panel Approach tovaluing productivity costs is accurate. The authors acknowledgethat using VAS as a valuation technique may have driven the

    results somewhat, as it may be expected to be relatively insensi-tive to income effects when respondents do not consider this apart of health.

    Krol and colleagues [35] attempt to replicate their first studynow using the TTO method. This method is not only betteraccepted as a valuation technique for health states, but alsoconceptually, it may allow respondents to more easily consider

    broader consequences of illness, like income changes, in theirvaluations. In this study, Krol and colleagues used the same threeversions of the questionnaire as above (no mention of income,explicit inclusion of income effects, and explicit exclusion ofthese effects), aim to test the same hypotheses and ask people tovalue the same three EQ-5D health states. A total of 210members of the general public in The Netherlands participated inthe study. For version 1, when there was no mention of incomeloss, 64% of respondents spontaneously included these effects (amuch higher value than previous studies), but as with the studyby Krol and colleagues [34] there were no significant differencesin valuations between the respondents who did and those whodid not spontaneously include income. When those that indicatedthat they had included income effects in version 1 in the ex antevaluation were asked to repeat the exercise assuming no income

    effects, the ex post valuation showed no significant differencescompared to the ex ante. Comparing responses to versions 2 and3, the finding of Krol and colleagues [32] were confirmed: that nosignificant differences between these versions occur. As with thestudy by Krol and colleagues [32] the average age of the samplewas quite low (35) and a large proportion did not have paidwork (28%), but again, within the experiment these characteris-tics cannot explain the results. The authors furthermore stressthat some results may have been insignificant due to a lack ofstatistical power, owing to the relatively small conveniencesample used (although this is the second largest sample of thestudies identified). Also, the TTO questions are framed in such amanner that may lead respondents to believe they can only makea minimum trade of 1 year. This is likely to effect the valuesobtained, especially in the case of the mildest state (21211), but

    again, it is unlikely that this has had an influence on the com-parisons made within this study.

    Richardson and colleagues [34] performed a study in Austra-lia, also using the TTO method. Individuals who had completedTTO interviews as part of the construction of the Vision Qualityof Life Index (VisQoL) multi-attribute utility instrument [38]were questioned about their assumptions concerning income andthe amount of thought given to income during a TTO interview.A total of 28 different multi-attribute health states were valued.The two hypotheses they aimed to test were: 1) when there is noexplicit statement about income in the exercise the majority ofTTO respondents will assume that in a very poor health statetheir incomes will fall; and 2) TTO scores will be lower whensubjects assume that their incomes will fall. The sample consistedof 70 visually impaired patients and 61 members of the general

    population. After completion of the TTO interview an incomequestionnaire was administered to the population group atfollow-up by post. Due to their visual impairment the question-naire was administered to the patient group as a face-to-faceinterview immediately after completion of their TTO questions.The questionnaire first asked respondents how much thoughtthey gave to income and spending, with possible answers of nothought, a little thought, or a lot of thought. Second, therespondents who had considered income and spending wereasked whether they assumed that their income would remain thesame, would be lower, or would be much lower.

    The results showed that only 9.1% of the respondents gave alot of thought to income and spending, while 62% gave the issue

    302 Tilling et al.

  • 7/29/2019 In r Out..Income Losses

    6/8

    no thought. While 38% of respondents did think about income alittle or a lot, only 26.7% of these believed their income wouldfall a little or a lot, leading to a rejection of the first hypothesis.Ordinary least squares regression was used to test the secondhypothesis. The results suggested a strong negative relationshipbetween thinking a lot about income and the disutility score (i.e.,the more the respondents thought about income the higher their

    TTO value, which is counterintuitive). However, the significanceof this result disappeared in the random effects model. Bothregressions only showed a statistically significant and lower dis-utility (i.e., a higher TTO score) associated with individuals whothought a lot about income and assumed it would decrease alittle. The authors conclude that this perverse result is almostcertainly attributable to the small number of respondents (7) inthis category. The most important conclusion of this study wasthat survey respondents do not appear to have reduced theirTTO estimates to take account of a possible loss of income whentheir health state implies an inability to work. Unfortunately,information on background characteristics was not presented. Itwould be particularly useful to have information on age andoccupation of the respondents to more incisively assess theresults obtained.

    A study was performed by Myers and colleagues [33]. Thiswas the only study using the standard gamble (SG) method ofpreference elicitation (for standard description of the SG methodsee Drummond et al. [6]). They administered a paper SG exerciseto 181 undergraduate economics students in the United States.The students were randomized into one of two groups. In group1 no information was provided regarding income effects, while ingroup 2 participants were simply informed that these effectsmight occur. Students were presented with one of three healthstates based on differing degrees of carpal tunnel syndrome (mild,moderate, and severe). These states were anchored to the loss ofboth hands rather than the conventional approach of anchoringto death. The results showed that the overall mean QoL for thegroup informed of the potential for income loss was lower thansubjects who were uninformed (P < 0.0001). For the mild state

    the mean valuation for the informed group was 0.832 comparedwith 0.903 for the uninformed group. For the severe state themeans are 0.626 for the informed group compared to 0.743 forthe uninformed group. Since differences were found between thetwo groups the authors concluded that income losses are notspontaneously included in the assessment of QoL weights.However, at the individual level it seems quite plausible thatsome may have spontaneously included these costs while othersmay not. Moreover, the mentioning of potential income loss mayhave increased the idea of severity of the projected health statefor the respondents. Background characteristics of the samplewere not presented, but since they were students they were pre-sumably relatively young and had relatively low incomes. Asindicated earlier, how this may have affected answers is difficultto predict. Also, since the health states were anchored to the loss

    of both hands rather than death, the comparability of the resultswith the other studies is questionable.

    Most recently, Davidson and Levin [36] asked 200 Swedishstudents to value four EQ-5D states (11211, 11122, 21232, and33321), through self-complete VAS and TTO exercises. The stu-dents were randomly allocated to one of two groups: either theincome or nonincome group. The nonincome group received noinstruction regarding income while completing the valuations.This group was asked a follow-up question: did you consideryour expected income when valuing the health states? Those thatstated that they had not included income effects were asked torevalue the states, this time including expected income (ex postinclusion). The income group received instructions to include

    effects on expected income. This group was told to assume aspecific gross income per month at full health. Four differentincome levels were used, but each student only had one incomelevel to consider.

    The background characteristics of the study show that thetwo groups did not differ significantly in any variable. In thenon-income group 94% of respondents did not spontaneously

    include income effects. When these were asked to revalue thestates including income effects, the TTO valuations were signifi-cantly lower for two of the four states, and the VAS valuationswere significantly lower for three of the four states. In the incomegroup, 40% believed that thinking about their expected incomehad affected their valuations of the health states. Students in theincome group were asked what percentage of their specifiedincome in full health they would expect to have in the healthstates. The mean expected income percentages in the four stateswere 73%, 82%, 47%, and 30%. Interestingly, although state11122 was valued lower than state 11211, respondents felt thatstates 11211 would have a larger effect on income. This suggeststhat the usual activities dimension is perceived to have the great-est effect on income. Comparison of the two groups showed thatthe income group gave significantly lower TTO valuations than

    the nonincome group for only the mildest state. There weresignificantly lower VAS valuations among the income group forthe two most severe states. The main weakness of this study isthat the sample consists of students. The authors argue thatstudents were used because they generally have low, and similar,incomes but with increasing expected incomes in the future.However, we would argue that students do not have mortgagesor dependents and hence cannot relate to the financial burdenand the stress associated with potentially not being able to meetthese demands. Given this, it is unsurprising that 94% of respon-dents did not spontaneously consider income effects. Theapproach of telling respondents to assume a fixed given incomein full health further complicates the exercises and increases thecognitive burden. Ultimately, the authors argue that productivitycosts should be included in the numerator rather than the

    denominator of a cost-effectiveness analysis.It is clear that the above studies draw inconsistent conclu-

    sions, use various preference elicitation methods, samples andstudy designs, and often suffer from important weaknesses.

    Discussion and Research Agenda

    While the Washington Panel recommendations may not havereceived much support, they suggest that the line between costsand effects in health economic evaluations has not been drawncarefully enough and, therefore, current methods may lead todouble-counting when used in combination. Besides the empiri-cal questions, a number of theoretical issues in this debate remainunresolved. First, it is dubious whether income effects can beconsidered a part of health effects, as suggested by the Washing-

    ton Panel. Second, the use of income as an accurate proxy forsocietal productivity costs is questionable. These considerationsmay already lead to a preference for including productivity costson the cost-side of the C/E ratio, but does not take away the needto address the empirical questions of whether, and how, peopletake account of income losses when valuing health states.

    The currently available empirical evidence on this issuecannot be considered decisive or conclusive. There are someinconsistencies between the conclusions of the existing studies.On the topic of spontaneous inclusion of income effects, Meltzerand colleagues [30] and Sendi and Brouwer [31] did find signifi-cant differences between valuations of respondents including/excluding income effects, while Krol and colleagues [32,35] and

    In or Out? 303

  • 7/29/2019 In r Out..Income Losses

    7/8

    Richardson and colleagues [34] did not. The lower VAS scoresfound by Sendi and Brouwer [31] among those who consideredincome effects could be caused by the fact that the sample wasvery small and consisted of health professionals. These are likelyto have a better understanding than the general public of theeffect a given health state will have on your ability to work andhence your income. Moreover, as in other studies, the results may

    also have been affected by differences in considering leisure timebetween respondents. It seems important to better control fordifferences in that respect. The studies on explicit ex ante instruc-tion regarding inclusion or exclusion of income (but withoutspecifying the size of the effect) are also inconclusive. Krol andcolleagues [32,35] found no significant differences. However,Davidson and Levine [36] found that respondents instructed toinclude income effects gave significantly lower valuations thanthose with no instructions in some cases. However, it seems thatexplicitly mentioning (potential) income losses (but not tellingrespondents what to do with this information) does affect HSV.The results presented by Meltzer and colleagues [30] for backpain (although not for blindness) seem to indicate this, as well asthe results of Myers and colleagues [33] for carpal tunnel syn-drome. Such results may also indicate that explicit mentioning of

    potential income losses may lead people to believe that the con-dition is more severe than they first imagined or that they empha-size other aspects of work (such as role functioning) in theirsubsequent valuation [24].

    Empirically, the clearest conclusion so far is that, withoutinstructions, some people do include income effects while othersdo not. The specific percentage of respondents that incorporatethese effects is less clear. Estimates range from 6% to 64% (whenincome is not mentioned). This indicates that whether productiv-ity costs are included in the numerator or the denominator, theymay be eitherunder or over accounted for, dependingon the effectof inclusion of income effects on HSVs. If productivity costs areincluded in the numerator andthe influence of income effects onHSVs is not negligible, then some double-counting will occur.Alternatively, if productivity costs are to be included in the

    denominator then productivity costs will be under accounted for,since some people will not consider income effects in their valua-tions (unless the influence on HSVs is unduly large). Furthermore,if HSVs would indeed contain non-negligible income effects, thisposes a problem when decision makers in certain jurisdictions donot wish to include productivity costs in health economic evalu-ations in any way. The evidence so far, however, suggests thatpossible double-counting, given the current way of deriving HSVs(e.g., without specifying the size of income effects) seems to benon-existent or, at least, negligible, given that most studies find nosignificant differences in values between those that do and do notconsider income effects when the measure is silent. Althoughexplicit instruction similarly seems to make little difference tovaluations, for now the best solution may be to explicitly excludeincome effects and include productivity costs in the numerator.

    The studies identified in this article have various weaknessesand, while answering some questions, have also resulted in newones. Most studies have a limited sample size and four out of theseven ask respondents to value specific health states (e.g., carpaltunnel syndrome) rather than generic health states. This is animportant difference as respondents may be able to imagine moreeasily the impact on their productivity due to specific diseasesthan due to a general health profile. In terms of relevance, itwould especially be useful to understand income effects withingeneric instruments as these are now the most commonly usedmetrics within economic evaluations and are also used to formwidely used population value sets. It is interesting to note that allstudies that have found significant differences in values have

    studied a specific condition. Neither of the studies using a genericinstrument, the EQ-5D, have found any notable differences. Thishighlights the possibility that respondents may be able to relateto a specific illness more easily and therefore may be morecapable and perhaps likely to envisage the broader consequencesof such an illness. It seems important to study the differencesbetween generic and disease-specific instruments further in future

    research. The evidence seems to suggest that population valuesets derived using generic instruments such as the EQ-5D are notnoticeably influenced or polluted by income effects and there-fore can be used alongside monetary valuation methods of pro-ductivity costs, as well as in contexts where income effects are tobe excluded from the analysis completely.

    Another noteworthy issue may be the experience people mayhave (themselves or in others) with some health state. Not onlymight this affect the valuation of that health state in general, butit may also raise awareness of the broader consequences of illnessand therefore may affect the inclusion of such consequences.Therefore, future research may study whether the influence ofknowledge of a specific health state or disease affects the likeli-hood of considering income effects in the valuation of that healthstate and, indeed, the valuation itself.

    Five of the seven studies reviewed, with the notable exceptionof the visually impaired patients in the Richardson and colleagues[34] study (while the mode of administration is not clear in thestudy by Meltzer andcolleagues [30]), use self-complete question-naires. Studies administering preference elicitation through inter-views, despite being more time consuming and expensive, wouldenable researchers to gain a greater understanding of the thoughtprocesses at work when respondents are answering the questions(perhaps through parallel qualitative probing).

    There is undoubtedly a need for further empirical testing withlarger sample sizes to further address the question of whatrespondents consider when answering preference elicitationquestions and how this influences their valuations. More inves-tigation is needed into the role of income effects in genericinstruments such as the one already used, the EQ-5D, but also

    others such as the SF-6D and the HUI. The HUI would beparticularly interesting as it already specifically rules out theconsideration of income effects. The effects of explicit instructionalso need to be investigated further. While it may be useful toenhance the consistency of the inclusion or exclusion of certaineffects by explicit instruction, it may have unintended side effectsas well. Explicit instruction to ignore certain, previously unmen-tioned, items from the valuation process may result in the oppo-site if people find it difficult to explicitly exclude items. Moreover,an explicit instruction to include certain aspects may lead them toreceive too much weight in the subsequent valuation procedure.Therefore, if further evidence shows that explicit instructionmakes no difference, then the exclusion of productivity costs inHSVs and thus from the numerator of the C/E ratio, might noteven need to be explicit. More research is needed in a wider

    selection of countries since results are likely to be very sensitive tothe local social security system.

    Some of the studies have not only looked at income effects butat leisure effects related to ill health as well (since impaired healthmay affect the utility derived from leisure time). It seems impor-tant to continue to do so in future studies, not only to find outwhether leisure is currently adequately valued in terms of QoL[39] but in addition to disentangle leisure and income effects inempirical studies.

    Finally, improved understanding may be gained through sepa-ration of the income and real health effects, e.g., by valuing justhealth through a TTO exercise (with explicit exclusion of incomeeffects) and then valuing only income through a TTO exercise. It

    304 Tilling et al.

  • 7/29/2019 In r Out..Income Losses

    8/8

    would be interesting to see if the relationship between incomeand health proves to be additive or if there is some, morecomplex, interaction between the two.

    In conclusion, it is far from clear whether income effects arein or out of HSVs and how this affects subsequent valuations.New studies with an adequate sample size are needed, preferablyusing representative samples of the general population in some

    jurisdiction. The use of generic health states seems to be mostinformative (since economic evaluations are now typicallyinformed by population value sets), such as EQ-5D states. Givenimportant differences between countries in terms of social secu-rity systems etc, consideration should be given to the generaliz-ability of results to other countries. Given the serious doubtsregarding the accuracy of valuing productivity losses throughQoL, monetary methods to include productivity costs seem pref-erable. Whether this requires explicit instructions in order toavoid double-counting needs further investigation. Until moreconclusive evidence is available in this area, economic evaluatorsseeking to include productivity costs may wish to present resultsfrom both the numerator and denominator methods as a sensi-tivity analysis. They may also consider presenting results withproductivity costs excluded altogether. This would highlight the

    influence on results, facilitate comparisons between jurisdictionsthat do and do not take a societal perspective, and help avoiddouble-counting. While it is too early to make a final statementon the question whether income is in or out of HSVs, the evi-dence so far largely suggests that in terms of a reliable andsubstantial impact on QoL it is out.

    Source of financial support: None.

    References

    1 Brouwer WBF, Koopmanschap MA, Rutten FHH. Productivitycosts measurement through quality of life? A response to therecommendation of the Washington panel. Health Econ1997;6:2539.

    2 Gold MR, Siegel JE, Russell LB, Weinstein MC. Cost-effectivenessin Health and Medicine. Oxford: Oxford University Press, 1996.3 Rothermich EA, Pathak DS. Productivity-cost controversies in

    cost-effectiveness analysis: review and research agenda. Clin Ther1999;21:25567.

    4 Pritchard C, Sculpher MJ. Productivity Costs: Principles andPractice in Economic Evaluation. London: Office of Health Eco-nomics, 2000.

    5 Tranmer JE, Guerriere DN, Ungar WJ, Coyte PC. Valuing patientand caregiver time. A review of the literature. Pharmacoeconom-ics 2005;23:44959.

    6 Drummond MF, OBrien B, Stoddart GL, Torrance GW.Methods for the Economic Evaluation of Health Care Pro-grammes (3rd ed.). New York: Oxford University Press, 2005.

    7 Brouwer WBF, Rutten FHH, Koopmanschap MA. Costing ineconomic evaluations. In: Drummond M, McGuire A, eds. Eco-nomic Evaluation in Health Care: Merging Theory with Practice

    (2nd edn). New York: Oxford University Press, 2005.8 Culyer AJ, Lavers R, Williams A. Social indicators: health. Social

    Trends 1971;2:3142.9 Weisbrod BA. The valuation of human capital. J Polit Econ 1961;

    69:42536.10 Rice D, Cooper B. The economic value of human life. Am J Public

    Health 1967;57:195466.11 Beutels P, Bonanni P, TormansG, et al. An economic evaluation of

    universal pertussis vaccination in Italy. Vaccine 1999;17:24009.12 Ford ES, Kelly AE, Teutsch SM, et al. Radon and lung cancer: a

    cost-effectiveness analysis. Am J Public Health 1999;89:3517.13 Kobelt G, Eberhardt K, Jonsson L, Jonsson B. Economic conse-

    quences of the progression of rheumatoid arthritis in Sweden.Arthritis Rheum 1999;42:34756.

    14 Koopmanschap MA, Rutten FHH, van Ineveld B, Van Roijen L.The friction cost method of measuring the indirect costs ofdisease. J Health Econ 1995;14:17189.

    15 Koopmanschap MA, van Ineveld B. Towards a new approach forestimating indirect costs of disease. Soc Sci Med 1992;34:100510.

    16 Koopmanschap M, Rutten FHH. Indirect costs in economicstudies. Pharmacoeconomics 1993;4:44654.

    17 Koopmanschap MA, Rutten FHH. Indirect costs: the conse-quence of production loss or increased costs of production. MedCare 1996;34(Suppl. 12):DS5968.

    18 Glick H, Willke R, Polsky D, et al. Economic analysis of tirilazadmesylate for aneurismal subarachnoid hemorrhage. Int J TechnolAssess Health Care 1998;14:14560.

    19 Van Enckevort PJ, TenVergert EM, Bonsel GJ, et al. Technologyassessment of the Dutch lung transplantation program. Int JTechnol Assess Health Care 1998;14:34456.

    20 Nord E, Wisloff F, Hjorth M, Westin J. Cost utility analysis ofmelphalan plus prednisone with or without interferon-alpha 2b innewly diagnosed multiple myeloma: results from a randomisedcontrolled trial. Pharmacoeconomics 1997;12:89103.

    21 Johannesson M, Karlson G. The friction cost method: a comment.J Health Econ 1997;6:24955.

    22 Liljas B. How to calculate indirect costs in economic evaluations.Pharmacoeconomics 1998;13(1 Pt 1):17.

    23 Koopmanschap M, Rutten F, van Ineveld B, Van Roijen L. Replyto Johannessons and Karlssons comment. J Health Econ 1997;16:25759.

    24 Johannesson M. Avoiding double-counting in pharmacoeconomicstudies. Pharmacoeconomics 1997;11:3858.

    25 Brouwer WBF, Meerding WJ, Lamers L, Severens H. Productivityand health related quality of lifean exploration. Pharmacoeco-nomics 2005;23:20918.

    26 Torrance GW, Sackett D, Thomas W. A utility maximising modelfor evaluation of health care programmes. Health Serv Res 1972;7:11833.

    27 Wang PS, Patrick A, Avorn J, et al. The costs and benefits ofenhanced depression care to employers. Arch Gen Psychiatry2006;63:134553.

    28 Brouwer WBF, Koopmanschap MA, Rutten FHH. Productivitycosts in cost-effectiveness analysis: numerator or denominator: a

    further discussion. Health Econ 1997b;6:51114.29 Meltzer D, Johannesson M. Inconsistencies in the Societal Per-spective on costs of the panel on cost-effectiveness in health andmedicine. Med Decis Making 1999;19:3717.

    30 Meltzer D, Weckerle CE, Chang LM. Do people consider finan-cial effects in answering quality of life questions? Med DecisMaking 1999;19:517.

    31 Sendi P, Brouwer WBF. Is silence golden? A test of the incorpo-ration of the effects of ill-health on income and leisure in healthstate valuations. Health Econ 2005;14:6437.

    32 Krol M, Brouwer WBF, Sendi P. Productivity costs in health-statevaluations: does explicit instruction matter? Pharmacoeconomics2006;24:40114.

    33 Myers J, McCabe S, Gohmann S. Quality-of-life assessment whenthere is a loss of income. Med Decis Making 2007;27:2733.

    34 Richardson J, Peacock S, Lezzi A. Do quality-adjusted life yearstake account of lost income? Evidence from an Australian survey.

    Eur J Health Econ 2009;10:1039.35 Krol M, SendiP, Brouwer WBF. Breaking thesilence:exploring the

    potential effects of explicit instructions on incorporating incomeand leisure in TTO exercises. Value Health 2009;12:17280.

    36 Davidson T, Levin L. Do individuals consider expected incomewhen valuing health states? Int J Technol Assess Health Care2008;24:48894.

    37 Brooks R. Euroqol: the current state of play. Health Policy1996;37:5372.

    38 Misajon R, Hawthorne G, Richardson J, et al. Vision and qualityof life: the development of a utility measure. Invest OphthalmolVis Sci 2005a;46:400715.

    39 Brouwer WBF, Rutten FHH. The missing link: on the linebetween C and E. Health Econ 2003;12:62936.

    In or Out? 305