UBC Faculty Research and Publications

Do visual analogue scale (VAS) derived standard gamble (SG) utilities agree with Health Utilities Index… Rashidi, AmirAdel; Anis, Aslam H; Marra, Carlo A Apr 20, 2006

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12955_2005_Article_259.pdf [ 537.51kB ]
JSON: 52383-1.0132614.json
JSON-LD: 52383-1.0132614-ld.json
RDF/XML (Pretty): 52383-1.0132614-rdf.xml
RDF/JSON: 52383-1.0132614-rdf.json
Turtle: 52383-1.0132614-turtle.txt
N-Triples: 52383-1.0132614-rdf-ntriples.txt
Original Record: 52383-1.0132614-source.json
Full Text

Full Text

ralHealth and Quality of Life OutcomesssBioMed CentOpen AcceResearchDo visual analogue scale (VAS) derived standard gamble (SG) utilities agree with Health Utilities Index utilities? A comparison of patient and community preferences for health status in rheumatoid arthritis patientsAmir Adel Rashidi1, Aslam H Anis2 and Carlo A Marra*3,4Address: 1Centre for Clinical Epidemiology and Evaluation, Faculty of Pharmaceutical Sciences, University of British Columbia, Canada, 2MHA Program, Department of Health Care and Epidemiology, Faculty of Medicine, University of British Columbia, Canada, 3Faculty of Pharmaceutical Sciences, University of British Columbia, Vancouver, BC, Canada and 4Centre for Clinical Epidemiology and Evaluation, Vancouver Coastal Health Research Institute, Vancouver, BC, CanadaEmail: Amir Adel Rashidi - amiradel@interchange.ubc.ca; Aslam H Anis - aslam.anis@ubc.ca; Carlo A Marra* - carlo.marra@ubc.ca* Corresponding author    AbstractBackground: Assessment of Health Related Quality of Life (HRQL) has become increasingly importantand various direct and indirect methods and instruments have been devised to measure it. In directmethods such as Visual Analog Scale (VAS) and Standard Gamble (SG), respondent both assesses andvalues health states therefore the final score reflects patient's preferences. In indirect methods such asmulti-attribute health status classification systems, the patient provides the assessment of a health stateand then a multi-attribute utility function is used for evaluation of the health state. Because these functionshave been estimated using valuations of general population, the final score reflects community'spreferences. The objective of this study is to assess the agreement between community preferencesderived from the Health Utilities Index Mark 2 (HUI2) and Mark 3 (HUI3) systems, and patientpreferences.Methods: Visual analog scale (VAS) and HUI scores were obtained from a sample of 320 rheumatoidarthritis patients. VAS scores were adjusted for end-aversion bias and transformed to standard gamble(SG) utility scores using 8 different power conversion formulas reported in other studies. Individual levelagreement between SG utilities and HUI2 and HUI3 utilities was assessed using the intraclass correlationcoefficient (ICC). Group level agreement was assessed by comparing group means using the paired t-test.Results: After examining all 8 different SG estimates, the ICC (95% confidence interval) between SG andHUI2 utilities ranged from 0.45 (0.36 to 0.54) to 0.55 (0.47 to 0.62). The ICC between SG and HUI3utilities ranged from 0.45 (0.35 to 0.53) to 0.57 (0.49 to 0.64). The mean differences between SG and HUI2utilities ranged from 0.10 (0.08 to 0.12) to 0.22 (0.20 to 0.24). The mean differences between SG and HUI3utilities ranged from 0.18 (0.16 to 0.2) to 0.28 (0.26 to 0.3).Conclusion: At the individual level, patient and community preferences show moderate to strongagreement, but at the group level they have clinically important and statistically significant differences.Using different sources of preference might alter clinical and policy decisions that are based on methodsPublished: 20 April 2006Health and Quality of Life Outcomes 2006, 4:25 doi:10.1186/1477-7525-4-25Received: 26 July 2005Accepted: 20 April 2006This article is available from: http://www.hqlo.com/content/4/1/25© 2006 Rashidi et al; licensee BioMed Central Ltd.This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.Page 1 of 10(page number not for citation purposes)that incorporate HRQL assessment. VAS-derived utility scores are not good substitutes for HUI scores.Health and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/25BackgroundIn recent years, cost-utility analysis has emerged as a com-mon methodology for the economic evaluation of healthcare strategies. This approach makes use of qualityadjusted life years (QALYs) to assess the effectiveness ofhealth care interventions. Neumann et al. stated that"QALYs represent the benefit of a health intervention interms of time in a series of quality-weighted health states"in which the quality weights reflect the desirability of liv-ing in the state [1]. Therefore, once the quality weights areobtained for each health state experienced by an individ-ual, they are multiplied by the duration of time spent inthe health state. The products of these calculations arethen summed to obtain the total number of QALYs.Preference-based assessments, which can be categorizedinto direct and indirect measures, are often used to obtainthe desirability or preferences for health states. In directmeasures, the respondent directly "assesses" and "evalu-ates" a health state on a scale of 0.00 (death) to 1.00 (per-fect health). The health states that are evaluated in thedirect approach can be hypothetical or can be therespondent's own subjectively defined current health state(SDCS) [2].In indirect measures, the respondent provides informa-tion regarding their health status by completing a multi-attribute health status classification system questionnairesuch as the Health Utilities Index Mark 2 (HUI2) [3] andMark 3 (HUI3) [4], the Quality of Well Being (QWB) [5],the EuroQol (EQ-5D) [6,7] and the Short-Form 6-D (SF-6D) [8]. The "valuation" of that assessment then comesfrom a scoring formula which is typically based on prefer-ences for health states from a general population sample.Direct methods include the visual analog scale (VAS), andstandard gamble (SG) techniques. The SG requiresrespondent's concentration, sound cognitive functioning,and requires experienced interviewers with effective props[9,10]. Since multi-attribute health status classificationsystem questionnaires can be self-administered, or com-pleted through telephone interviews, they have been morewidely used.Alternatively, some researchers have tried to use simpleindirect techniques such as the VAS and then convertedthe scores to SG utilities using power transformations[11,12].Although different variations of VAS have been frequentlyused as a simple method of preference measurement,recently some concerns regarding their validity have beenraised [13-15]. For example, the VAS anchors are often notis evidence that limited and cautious use of the VAS is use-ful and appropriate [16].Different approaches, considering preferences of differentpopulation subgroups, have been used to elicit the "val-ues" of various health states [17]. However, the two mainsources of values are individual patients and the generalpopulation. On one hand, it is felt that patients who havedirectly experienced a health state can better assess itseffect on their HRQL and express a true preference. On theother hand, members of the general public are less likelyto have self-interest or strategic bias in their evaluationsand thus may be more objective. Moreover, since the gen-eral public incurs the cost of resource allocation decisions,it may be more reasonable to measure preferences forhealth states and benefits from the general public's per-spective [17].Currently, economic evaluation guidelines recommendusing preference-based valuation methods in which thegeneral public is the source of values [18,19]. However, itis not clear whether community members value a givenhealth state the same as patients who are experiencing thathealth state. If there are significant differences betweenthese, then the results of economic evaluations couldchange depending on the preference source. Althoughseveral studies have shown that patient-based and com-munity-based utilities are significantly different [10,20-22], some other studies have shown otherwise [23,24].Recently, Feeny and colleagues reported differencesbetween utilities derived from the HUI2 and SG at theindividual level, but at the same time observed no differ-ence at the group level [2,25].As such, our objective was to assess the agreementbetween indirectly obtained community preferences anddirectly obtained patient preferences in a sample of rheu-matoid arthritis patients.MethodsStudy sampleA sample of patients with a rheumatologist-confirmeddiagnosis of rheumatoid arthritis (RA) was previouslyassembled for a longitudinal study to examine the relia-bility and responsiveness of the indirect utility instru-ments [26-28]. All participants provided informedconsent and ethical approval for this study was obtainedthrough the University of British Columbia's BehaviouralEthics Committee. Three hundred and twenty patientstook part in the study and data were gathered at threeintervals: baseline (Assessment A), after 3 months (Assess-ment B) and after 6 months (Assessment C).Page 2 of 10(page number not for citation purposes)well defined and several measurement biases such as con-text bias and end-aversion bias may occur. However, thereHealth and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/25Indirect and direct assessment of preferences for health statesThe study questionnaire included the HUI Mark 2 and 3,and the EQ-5D. Patients' preferences for their currenthealth state were obtained using a VAS as part of the EQ-5D questionnaire. The EQ-5D questionnaire [6,7] consistsof a descriptive health profile including five domains anda health thermometer (VAS) which represents a subjec-tive, global evaluation of the respondent's health statuson a vertical scale between 0 and 100, where 0 (the bot-tom anchor) represents the worst imaginable health stateand 100 (the top anchor) represents the best imaginablehealth state.Adjustment for end-aversion biasMany respondents are unwilling to place health states atthe extreme portions of a continuous scale, leading toend-aversion bias [29,30]. The magnitude of end-aversionbias in VAS has been investigated using the pair-wise com-parison method [16,31]. It was found that, on average,health states close to the healthy end are placed 1.78 timestoo far away, whereas at the unhealthy end, there is mini-mal bias. As such, only VAS scores placed in the upperquarter of the scale were adjusted and, in order to main-tain the relative position of other scores, a positive lineartransformation was performed. No adjustment was per-formed for the unhealthy end (closer to zero). This proce-dure is similar to the adjustment method performed indevelopment of HUI3 [4].Transformation of VAS scores to utility scoresUtilities for the respondent's SDCS were derived using atransformation function to convert adjusted VAS values(V) to SG utility scores (U). After adjustment for end-aver-sion bias, VAS scores first were transformed from a 0–100scale to a 0.00–1.00 scale. Then, power functions wereused to transform the data to SG utility scores. Power con-version is the most common transformation functionused for mapping the relationship between VAS scoresand SG utilities [16]. All eight different functions, previ-ously described by Torrance [16], were used to performthe transformations (Table 1).HUI2 and HUI3Each HUI system includes a health status classificationsystem and a multi-attribute utility scoring formula. TheHUI2 consists of questions regarding seven dimensions ofhealth status: sensation, mobility, emotion, cognition,self-care, pain, and fertility. Because each questiondescribes 3 to 5 levels of a health attribute, the HUI2 candescribe a total of 24,000 unique health states [3]. TheHUI3 consists of questions regarding eight dimensions ofhealth status: vision, hearing, speech, ambulation, dexter-ity, emotion, cognition, and pain. Because each questiondescribes 5 to 6 levels of a health attribute, the HUI3 candescribe a total of 972,000 unique health states [4]. Themulti-attribute utility scoring formula calculates a utilityscore that reflects community preferences for the respond-ent's assessment of his or her health status. The scoringformulae are based on SG utilities derived mainly frompower conversions of VAS scores. The overall utility scoresobtained from HUI2 range from -0.03 to 1.0 and for HUI3from -0.36 to 1.0, where 1.0 represents a HRQL of perfecthealth and 0 represents a HRQL of death. However, theoverall utility scores for HUI 2 and HUI3 can also be cal-culated such that 0.00 represents the worst imaginablehealth state and 1.00 represents the perfect health [3,4].Statistical analysisThe HUI2 and HUI3 scores were considered indirect com-munity-preference-based utility scores. VAS scores wereadjusted for end-aversion bias, and after conversion to SGutility scores were considered direct patient-preference-based utility scores (adjusted SG utility). SG utility scoreswere also calculated without adjusting for end-aversionbias (unadjusted SG utility). Both adjusted and unad-justed SG utility scores were calculated using each of theeight power conversion formulae (Table 1).VAS values (and therefore the obtained SG utility scores)are bound between 0.00 and 1.00. In order to avoid com-paring agreement between two utility measures with dis-similar ranges, the HUI2 and HUI3 scores were calculatedin a 0.00 to 1.00 scale in this study.Table 1: Different power functions reported for transforming VAS values (V) to SG utilities (U)*Function number Equation Reference1 U = 1-(1-V)1.6 Torrance et al.[51]2 U = 1-(1-V)2.2 Wolfson et al.[52]3 U = 1-(1-V)2.3 Torrance et al.[3]4 U = 1-(1-V)2.4 Feeny et al.[53]5 U = 1-(1-V)2.7 Krabbe et al.[54]6 U = 1-(1-V)2.9 Feeny et al.[53]7 U = V0.56 Furlong et al.[55]8 U = V0.47 Furlong et al.[55] and Le Gales et al.[56]Page 3 of 10(page number not for citation purposes)*Obtained from Torrance [16]Health and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/25Descriptive statistics are presented for each set of utilityscores. Agreement between SG utility scores and HUI2and HUI3 scores, at the individual level, was assessedusing the Pearson Correlation Coefficient and the Intrac-lass Correlation Coefficient (ICC) with a two-way mixedeffect model such that the respondent effect was randomand the measure effect was fixed [32]. Both the adjustedand unadjusted SG utility scores were examined sepa-rately. Interpretation of the strength of agreement usingICC scores was taken from the framework proposed byGuyatt et al. (strong: ICC>0.50; moderate: ICC = 0.35–0.50; weak: ICC = 0.20–0.34; negligible: ICC = 0.00–0.19)[33]. Paired sample t-tests were used to assess agreementbetween direct and indirect utility scores at the grouplevel. All the above tests were performed to assess agree-ment between the HUI scores and each SG utility scorecalculated from the different power conversions (8adjusted and 8 unadjusted). The minimal important dif-ference (MID) of utilities was considered to be 0.03 [9].A 0.05 level of significance was used in all analyses. ICCanalyses were carried out using SPSS version 11.5. Allother statistical analyses were performed using SAS ver-sion 8.2.ResultsRespondentsFrom the 320 participants who received the baseline ques-tionnaire (Assessment A), 308 completed the VAS scoresas part of EQ-5D questionnaire, and 307 and 306 globalutility scores could be generated using HUI scoring func-tions for the HUI2 and HUI3, respectively. Of these, 303respondents had both VAS and HUI2 scores and 302 hadboth VAS and HUI3 scores. Summary statistics for theeight different SG scores derived from VASs and HUI2 andHUI3 scores are presented in Table 2. More informationregarding the demographic characteristics and diseaseseverity of the study population has been published else-where [27,28].Individual level agreement between direct and indirect utilitiesIndividual level ICCs and Pearson correlation coefficientswere calculated where all 3 scores (VAS, HUI2 and HUI3)were available. The complete ICC analysis of AssessmentA along with the Pearson correlation coefficients is pre-sented in Table 3. In general, based on ICC results, mod-erate to strong agreement was found between SG utilitiesand HUI2 and HUI3 utilities at the individual level.The ICCs (95% confidence interval) between the adjustedSG and HUI2 utilities in Assessment A ranged from 0.45(0.36 to 0.54) to 0.55 (0.47 to 0.62), where most ICCsthe corresponding adjusted SG and HUI2 utilities with noICC below 0.50. These results show that agreementbetween the SG and HUI2 scores at the individual level isstrong. However, there is only moderate agreement at theindividual level between the SG and HUI3 utilities. TheICC (95% confidence interval) between the adjusted SGand HUI3 utilities in Assessment A ranged from 0.45(0.35 to 0.53) to 0.57 (0.49 to 0.64). ICCs between theunadjusted SG and HUI3 utilities were all higher than theICCs between the corresponding adjusted SG and HUI3utilities. In almost all measurements, the Pearson correla-tion coefficients slightly exceeded the correspondingICCs. However, none of the differences were statisticallysignificant. The analyses of Assessments B and C com-pletely support these findings (data not shown).Group level agreement between direct and indirect utilitiesResults of the comparison between the mean SG utilities,HUI2, and HUI3 scores using paired sample t-tests arereported in Table 4. The differences between the SG utili-ties and the HUI scores (the HUI score was subtractedfrom the SG utility) were calculated for every respondentand then the mean of the differences was examined forstatistical significance and clinical importance.In general, the mean differences between the SG utilitiesand HUI2 and HUI3 scores were important and statisti-cally significant. They were all positive, showing that theSG utilities consistently exceeded HUI utilities. The meandifferences between adjusted SG utilities and HUI2 scoreswere considerable but not so large. The mean (95% confi-dence interval) ranged from 0.10 (0.08 to 0.12) to 0.22(0.20 to 0.24). The mean differences between the adjustedSG utilities and HUI3 scores were larger, ranging from0.18 (0.16 to 0.20) to 0.28 (0.26 to 0.30).As expected, the mean differences between the unadjustedSG utilities and HUI2 scores were all smaller than themean differences between the corresponding adjusted SGutilities and HUI2 scores, but all were important and sta-tistically significant. The same was true for HUI3 scores.Analysis of Assessments B and C showed the same results(data not shown).DiscussionOur results indicate that at the individual level, goodagreement exists between SG and HUI utility scores. Theagreement between SG and both HUI2 and HUI3 utilitiesis generally strong (ICC>0.50). Also, at the group level wefound that SG and HUI utilities have important and sig-nificant differences. The differences were relatively largeand systematically in the same direction. Interestingly,our findings are in contrast with the results from Feeny etPage 4 of 10(page number not for citation purposes)were more than 0.50. ICCs between the unadjusted SGand HUI2 utilities were all higher than the ICCs betweenal. [2,25] and others [21,34,35].Health and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/25Agreement between direct and indirect utilities at the individual levelWhy is agreement less than perfect? How can we explainthe approximately 50 percent disagreement betweendirect and indirect utilities? And what are the possiblesources of disagreement between these utilities?The first explanation could be that direct and indirect util-ities measure preferences for health states from differentperspectives. While SG and HUI scores are both utilities,in direct measurement (SG), patient preferences are thebasis of the health status valuation, whereas in indirectassessment (HUI), the valuation is based on communitypreferences. In the direct SG measurement of a patient'scurrent health state, the patient makes a subjective assess-ment of his or her health status and then gives his or herpersonal evaluation of that health state. However, inmulti-attribute health status classification systems, such asTable 2: Summary statistics for HUI2, HUI3 and SG utilities obtained from transformation of VAS scores by different power conversionsAssessment A N Mean SD Median Min. Max.Unadjusted SG11 308 0.79 0.18 0.85 0.08 1.00SG2 308 0.86 0.16 0.93 0.11 1.00SG3 308 0.87 0.15 0.93 0.11 1.00SG4 308 0.88 0.15 0.94 0.12 1.00SG5 308 0.90 0.14 0.96 0.13 1.00SG6 308 0.91 0.13 0.97 0.14 1.00SG7 308 0.77 0.15 0.81 0.18 0.99SG8 308 0.80 0.13 0.84 0.24 1.00Adjusted SG1 308 0.84 0.18 0.92 0.09 1.00SG2 308 0.90 0.15 0.97 0.12 1.00SG3 308 0.91 0.14 0.97 0.13 1.00SG4 308 0.91 0.14 0.98 0.13 1.00SG5 308 0.93 0.13 0.99 0.15 1.00SG6 308 0.93 0.12 0.99 0.16 1.00SG7 308 0.82 0.15 0.88 0.19 1.00SG8 308 0.85 0.13 0.90 0.26 1.00HUI2 307 0.72 0.19 0.75 0.12 1.00HUI3 306 0.66 0.21 0.68 0.14 1.001Numbers indicate the power conversions (listed in Table 1) used to transform VAS scores to SG scores.Table 3: Pearson (r) and Intraclass (ICC) correlation coefficients between eight different SG scores (both adjusted and unadjusted) and HUI2 and HUI3. The 95% confidence intervals for ICCs are includedSG utility HUI2 HUI3Assessment A r ICC 95% CI r ICC 95% CIUnadjusted SG1 60% 0.57 0.49 to 0.64 60% 0.60 0.52 to 0.66SG2 55% 0.54 0.46 to 0.62 58% 0.56 0.47 to 0.63SG3 55% 0.54 0.45 to 0.61 58% 0.55 0.47 to 0.62SG4 55% 0.53 0.45 to 0.61 57% 0.54 0.46 to 0.62SG5 54% 0.51 0.42 to 0.59 56% 0.52 0.43 to 0.60SG6 53% 0.50 0.41 to 0.58 55% 0.50 0.41 to 0.58SG7 58% 0.60 0.48 to 0.63 62% 0.58 0.49 to 0.68SG8 58% 0.53 0.45 to 0.61 61% 0.54 0.46 to 0.62Adjusted SG1 55% 0.55 0.47 to 0.62 58% 0.57 0.49 to 0.64SG2 53% 0.51 0.42 to 0.59 55% 0.52 0.43 to 0.60SG3 53% 0.51 0.42 to 0.58 55% 0.51 0.42 to 0.59SG4 52% 0.50 0.41 to 0.58 55% 0.50 0.41 to 0.58SG5 51% 0.47 0.38 to 0.56 53% 0.47 0.38 to 0.55SG6 50% 0.45 0.36 to 0.54 52% 0.45 0.35 to 0.53SG7 57% 0.55 0.47 to 0.62 60% 0.56 0.48 to 0.64Page 5 of 10(page number not for citation purposes)SG8 57% 0.53 0.44 to 0.60 60% 0.53 0.44 to 0.61Health and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/25the HUI2 and HUI3, the patient provides the assessmentof his or her health state and then a multi-attribute utilityfunction (which has been estimated using the preferencesof general population) is used to evaluate the health state[25].This difference in perspective might lead to unequalresults for utility measurements which can be explainedby a phenomenon called response shift. Response shiftoccurs when the meaning of one's self-evaluation changes[36]. In general, patients who have experienced a chronichealth condition, such as RA, may give that health state ahigher value compared to the general public. Healthyindividuals might have an exaggerated fear of the morbid-ity and disability associated with such a chronic illnesses,while chronically ill patients often learn how to cope withtheir condition over time. Specifically, studies of rheu-matic diseases have shown that patients' self-reportedfunctional limitation and their actual physical impair-ment are considerably different [37]. Response shift mayoccur because of a change in the respondent's internalstandards of measurement (scale recalibration) [38], con-ceptualization of the health condition (concept redefini-tion) [39], or values [40].Another explanation for disagreement between direct andindirect utilities might reside in the selection of specificfunctional domains within HUI systems and the way thedomains are combined to generate a multi-attribute util-ity function. In the HUI systems, similar to many genericquestionnaires designed to evaluate quality of life, no dis-ease label is attached and only few aspects that determinehowever, the individual evaluates his or her own healthstate based on a holistic concept and determines a globalvalue for a global notion that includes not only his or herlevel of functioning but also the diagnosis, probable out-comes, and available treatment options. In addition tothis, one individual might value a domain, such as mobil-ity, twice as much as a different domain, such as cogni-tion. Another person might value it only half as much. Inindirect measures, the multi-attribute utility functiongives a single global assessment score for the HRQL,thereby suppressing the interpersonal heterogeneity inpreferences for domains. Direct measures, however,reflect this heterogeneity [41,42]. Some studies havefound that, for the majority of individuals, incorporatingthe relative importance of domains in indirect HRQLmeasurement has little effect on the accuracy of utilityestimation [43]. While this means that consideration ofrelative domain preferences does not significantly changethe results at the group level, as the authors confirmed, itmight be important at the individual level of analysis.Another source of disagreement could stem from themethod we used to obtain SG "utilities" from VAS "val-ues". VAS and SG techniques both quantify preferences;however, since their measurement approach is different,there is an essential dissimilarity between their scores. Inhealth status assessment, the subject is asked to comparetwo or more health states and then make a choice betweenthem or scale the alternatives. In the VAS technique, thequestion is framed under certainty, thus VAS is regarded asa measurable value function and represents the strength ofpreference under certainty. In contrast, in the SG tech-Table 4: Results of the comparison between mean SG utilities and HUI2 and HUI3 scores using paired sample t-testsSG utility HUI2 HUI3Assessment A N Mean Difference 95% CI N Mean Difference 95% CIUnadjusted SG1 303 0.07 0.05 to 0.09 302 0.13 0.11 to 0.15SG2 303 0.14 0.12 to 0.16 302 0.20 0.18 to 0.22SG3 303 0.15 0.13 to 0.17 302 0.21 0.19 to 0.23SG4 303 0.16 0.14 to 0.18 302 0.22 0.20 to 0.24SG5 303 0.18 0.16 to 0.20 302 0.24 0.22 to 0.26SG6 303 0.19 0.17 to 0.21 302 0.25 0.23 to 0.27SG7 303 0.05 0.03 to 0.07 302 0.11 0.09 to 0.13SG8 303 0.09 0.07 to 0.11 302 0.15 0.13 to 0.17Adjusted SG1 303 0.12 0.11 to 0.14 302 0.18 0.16 to 0.20SG2 303 0.18 0.16 to 0.20 302 0.24 0.22 to 0.26SG3 303 0.19 0.17 to 0.21 302 0.25 0.23 to 0.27SG4 303 0.19 0.18 to 0.21 302 0.25 0.23 to 0.27SG5 303 0.21 0.19 to 0.23 302 0.27 0.25 to 0.29SG6 303 0.22 0.20 to 0.24 302 0.28 0.26 to 0.30SG7 303 0.10 0.08 to 0.12 302 0.16 0.14 to 0.18SG8 303 0.13 0.11 to 0.15 302 0.19 0.17 to 0.21Page 6 of 10(page number not for citation purposes)quality of life of an individual are captured and summa-rized as a global score. In VAS and SG valuation methods,nique, which is based on the expected utility theory axi-oms [9,44-46], the question is framed under uncertainty,Health and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/25thus SG is considered as a utility function and representsthe strength of preference under uncertainty [16]. As aresult, SG "utilities" convey some extra information aboutthe subject's risk attitude which is not included in VAS"values". Dyer and Sarin [47] named this extra informa-tion as "relative risk attitude" which is different from theconventional concept of risk attitude. These authorsexplained that as the quantity of risky alternatives isincreased or decreased, the marginal value of additionalunits of those risky alternatives might change and that thischange in marginal value should be separated from peo-ple's attitude toward risk. They suggested that an individ-ual's relative risk attitude might be independent of theattribute on which his or her preferences are assessed andconsequently proposed that it might be appropriate toobtain "values" and then transform them to "utilities"using a relative risk attitude obtained from others whorepresent the decision maker [47]. Based on the consistentobservation that VAS values are lower than SG utilities,and that both scores are anchored at dead = 0.00 andhealthy = 1.00, Torrance and colleagues concluded that ifthere is a systematic relationship between the two meas-ures, it should be a concave curve that passes through 0and 1 [16]. They determined that a power conversionfunction fulfils these criteria.In order to test whether the effect of power conversionmight help explain the lack of perfect agreement betweendirect and indirect utility measurements, we also assessedthe agreement between VAS and HUI scores and com-pared them with ICCs between SG and HUI scores (resultsnot shown). In all three assessments (A, B and C) and forboth HUI2 and HUI3, transformation of VAS values to SGutilities decreased the agreement. Better agreementbetween rating scales and HUI scores than between SGand HUI scores has also been noted by Bosch et al. [48] ina study conducted on patients with intermittent claudica-tion. These results support the claim that power conver-sion might not be the best function to transform VASvalues to SG utilities. Other studies have examined therelationship between values and utilities and were unableto confirm the power function with their data [49]. How-ever, even though the appropriateness of using powerconversion to transform VAS values to utility scores isuncertain, we believe this factor has not significantly con-tributed to the observed disagreement. We calculatedPearson coefficients as well as ICCs in our analysis (Table3). Pearson coefficient only examines how well the rank-ing of health states from the best to the worst are compa-rable between SG and HUI. In the ICC method on theother hand, the absolute values of utilities are taken intoaccount. Therefore it is reasonable to expect that Pearsoncoefficients will be greater than ICC values. Comparisongreater than the corresponding ICC. However, the magni-tudes of the differences were negligible (maximum 7%)and none of them were statistically significant. Thereforewe expect factors, other than power conversion, to beresponsible for the detected disagreement. It is worthreminding that in development of the HUI2 and HUI3systems, the same method (power conversion) was usedto estimate SG utilities [3,4], therefore whatever the effectof power conversion is, it is common between the SG util-ities calculated in this study and HUI scores obtainedfrom scoring formulas in our study. However, our resultswere consistent across several power functions (Table 3).Interestingly, the smallest ICC was consistently obtainedusing the same power function as has been used to gener-ate the HUI2.Agreement between direct and indirect utilities at the group levelAt the group level, direct and indirect utilities showedimportant and statistically significant differences. How-ever, after observing strong agreement at the individuallevel, we expected otherwise. This is because direct meas-ures preserve individual variability in utility scores,whereas in the scoring formulas of HUI systems, individ-ual utilities are averaged and this variability is suppressed.One explanation for disagreement at the group level is theconcept of response shift, as discussed above. If we agreethat chronically ill patients usually become accustomed totheir situation, patient and community utilities shouldnot match and patient utilities should exceed those of thecommunity. This argument is supported by our findingsbecause, regardless of the effect of adjustment, theobserved differences in our t-test analysis are consistentlypositive in all eight power functions and three assess-ments.Although our analysis demonstrated obvious differencesbetween the two HUI systems, we did not intend to com-pare HUI2 and HUI3 systems in this study. Similar rela-tionship between HUI2 and HUI3 scores has beenreported and possible explanations for such differenceshave been presented elsewhere [4,25,27,28].Study limitationsIn measuring preferences for health states, a predefinedhypothetical health state can be explained to the respond-ent. Alternatively, the subject can be asked to evaluate hisor her own SDCS [2]. In this study, VAS scores wereobtained from patients with their SDCS in mind. If weassume that a respondent's conceptualization of healthstatus included some other dimensions not included inthe HUI2 and HUI3 systems, then in this study we haveactually compared different health states to each other.Page 7 of 10(page number not for citation purposes)of the Pearson correlation coefficients and ICCs showedthat in almost all assessments, the Pearson coefficient wasThis limitation might explain at least some part of theHealth and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/25observed disagreement between direct and indirect utili-ties.A power conversion specific to this study was not esti-mated. It seems that individuals do not have a context-independent relative risk attitude and a single power con-version can not be found to convert VAS scores to SGscores [15]. Torrance et al. explained that although con-text biases have been identified in several studies, the rela-tionship between VAS scores and SG utilities can bemodelled by a power curve specific to the study [16]. Theyemphasize that the power function should be developedwithin the same study. In development of the HUI2 andHUI3 systems, VAS scores and SG utilities were measuredfor a limited number of health states in the same study toestimate the power function which was used to transformthe scores. However, there are other studies that have notestimated their power function within the context of thatstudy and applied a power function reported by others[11,12]. Although this limitation could have affected theresults of current study, several power conversions wereexamined to minimize this shortcoming and the resultswere robust to utilization of various power functions.VAS measurements have several problems. First, if the topand bottom anchors of VAS are not clearly defined (e.g.dead), comparison of scores between individuals mightbe invalid. The anchors for the VAS used in this study (asincluded in the EQ5D questionnaire) were labeled "bestimaginable health state" and "worst imaginable healthstate". Clearly, these anchors can be conceptualized byindividuals differently. However, on the VAS used todevelop the HUI systems, the anchors were also labeled"best desirable" and "worst desirable" and were notclearly defined. Furthermore, VAS measurements areprone to several measurement biases such as spacing-outbias, end-aversion bias, and context biases [13,15]. In thisstudy, the effect of end-aversion bias at the upper end ofthe scale has been adjusted. However, there are othertypes of adjustment that could have been used to improvethe results, such as Parducci and Wedell's range-frequencymodel [50].ConclusionNational guidelines in Canada and the United States haverecommended using community-preference-based valua-tion methods, such as the HUI systems, for economicevaluations and HRQL assessments [18,19]. Due to thesimplicity of VAS measurements for both respondents andresearchers, there might be a tendency to measure patientpreferences using a VAS, adjust for biases, and then con-vert the scores to utilities using a power transformationfunction. Our study showed that for group level analysis,Furthermore, our results support the existence of responseshift phenomenon in chronically ill patients, explainingwhy patients usually give higher utility scores to their con-dition compared to the general public. This mightincrease the incremental cost-effectiveness ratio for somepreventive health interventions performed from thepatient's perspective compared to community's perspec-tive. Consequently, resource allocation decisions and theselection of health interventions for funding might greatlydepend on the source of preferences or on the assessmenttechnique.More research is needed to assess the agreement betweendirect and indirect preference measurement methods atthe individual and group levels.Authors' contributionsAAR participated in the design of the study, performed thebackground research, carried out the data analysis andinterpretation, and wrote the manuscript. AHA partici-pated in the design of the study and supervised theresearch activities. CAM participated in the design of thestudy, statistical analysis, interpretation of the results, andwriting the manuscript. All authors read and approved thefinal manuscript.AcknowledgementsThe authors would like to thank Ms. Megan Coombes for kindly reviewing and editing this paper. This work was supported by a grant from the Cana-dian Arthritis Network (a National Centre of Excellence). Dr. Marra is sup-ported by a Canadian Arthritis Network Scholar Award, and a Michael Smith Foundation for Health Research Scholar Award.References1. Neumann PJ, Goldie SJ, Weinstein MC: Preference-based meas-ures in economic evaluation in health care.  Annu Rev PublicHealth 2000, 21:587-611.2. Feeny D, Furlong W, Saigalf S, Sun J: Comparing directly meas-ured standard gamble scores to HUI2 and HUI3 utilityscores: group- and individual-level comparisons.  Soc Sci Med2004, 58:799-809.3. Torrance GW, Feeny DH, Furlong WJ, Barr RD, Zhang Y, Wang Q:Multi-attribute preference functions for a comprehensivehealth status classification system: Health utilities indexmark 2.  Med Care 1996, 34:702-722.4. Feeny DH, Furlong WJ, Torrance GW, Goldsmith CH, Zhu Z, DeP-auw S, Denton M, Boyle M: Multi-attribute and single-attributeutility functions for the health utilities index mark 3 system.Med Care 2002, 40:113-128.5. Patrick DL, Bush J, Chen M: Methods for measuring levels ofwell-being for a health status index.  Health Serv Res 1973, 8:228-245.6. Essink-Bot ML, Stouthard MEA, Bonsel GJ: Generalizability of val-uations on health states collected with the EuroQol Ques-tionnaire.  Health Econ 1993, 2:237-246.7. Rabin R, De Charro F: EQ-5D: A measure of health status fromthe Euroqol group.  Ann Med 2001, 33:337-343.8. Brazier J, Roberts J, Deverill M: The estimation of a preference-based measure of health from the SF-36.  J Health Econ 2002,21:271-92.9. Drummond MF, O'Brien B, Stoddart GL, Torrance GW: MethodsPage 8 of 10(page number not for citation purposes)VAS-derived utility scores are not good substitutes for HUIscores.for the economic evaluation of health care programmes.2nd edition. Oxford: Oxford Medical Publications; 1997. Health and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/2510. Furlong W, Feeny D, Torrance GW, Barr R, Horsman J: Guide todesign and development of health-state utility instrumenta-tion.  McMaster University Centre for Health Economics and PolicyAnalysis Working Paper; 1990:90-99. 11. Schackman BR, Goldie SJ, Freedberg KA, Losina E, Brazier J, Wein-stein MC: Comparison of health state utilities using commu-nity and patient preference weights derived from a survey ofpatients with HIV/AIDS.  Med Decis Making 2002, 22:27-38.12. Raat H, Bonsel GJ, Hoogeveen C, Essink-Bot ML, Dutch HUI Group:Feasibility and reliability of a mailed questionnaire to obtainvisual analogue scale valuations for health states defined bythe Health Utilities Index Mark 3.  Med Care 2004, 42:13-18.13. Bleichrodt H, Johannesson M: An experimental test of a theoret-ical foundation for rating scale valuations.  Med Decis Making1997, 17:208-216.14. Schwartz A: Rating scales in context.  Med Decis Making 1998, 18:236.15. Robinson A, Loomes G, Jones-Lee M: Visual analog scales, stand-ard gambles and relative risk aversion.  Med Decis Making 2001,21:17-27.16. Torrance GW, Feeny D, Furlong W: Visual analog scales: do theyhave a role in the measurement of preferences for healthstates?  Med Decis Making 2001, 21:329-334.17. Dolan P: Whose preferences count?  Med Decis Making 1999, 19:482-486.18. Canadian Coordinating Office for Health TechnologyAssessment: Guidelines for economic evaluation of pharma-ceuticals.  2 1997.19. Gold MR, Siegel JE, Russell LB, Weinstein MC: Cost-effectivenessin health and medicine.  New York: Oxford University Press;1996. 20. Postulart D, Adang EM: Response shift and adaptation in chron-ically ill patients.  Med Decis Making 2000, 20:186-193.21. Gabriel SE, Kneeland TS, Melton LJ III, Moncur MM, Ettinger B, Toste-son AN: Health-related quality of life in economic evaluationsfor osteoporosis: whose values should we use?  Med DecisMaking 1999, 19:141-148.22. Boyd NF, Sutherland HJ, Heasman KZ, Tritchler DL, Cummings BJ:Whose utilities for decision analysis?  Med Decis Making 1990,10:58-67.23. Llewellyn TH, Sutherland HJ, Tibshirani R, Ciampi A, Till JE, Boyd NF:Describing health states: methodologic issues in obtainingvalues for health states.  Med Care 1984, 22:543-552.24. Jenkinson C, Gray A, Doll H, Lawrence K, Keoghane S, Layte R: Eval-uation of index and profile measures of health status in a ran-domized controlled trial: comparison of the MedicalOutcomes Study 36-Item Short Form Health Survey, Euro-Qol, and disease specific measures.  Med Care 1997, 35:1109-1118.25. Feeny D, Blanchard C, Mahon JL, Bourne R, Rorabeck C, Stitt L, Web-ster-Bogaert S: Comparing Community preference-based anddirect standard gamble utility scores: evidence from electivetotal hip arthroplasty.  Intl J Tech Ass Health Care 2003, 19:362-372.26. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, CooperNS, Healey LA, Kaplan SR, Liang MH, Luthra HS, Medsger TA, Mitch-ell DM, Neustadt DH, Pinals RS, Schaller JG, Sharp JT, Wilder RL,Hunder GG: The American rheumatism Association 1987revised criteria for the classification of rheumatoid arthritis.  Arthritis Rheum 1988, 31:315-324.27. Marra CA, Woolcott JC, Kopec JA, Shojania K, Offer R, Brazier JE,Esdaile JM, Anis AH: A comparison of generic, indirect utilitymeasures (the HUI2, HUI3, SF-6D, and the EQ-5D) and dis-ease-specific instruments (the RAQoL and the HAQ) inrheumatoid arthritis.  Soc Sci Med 2005, 60:1571-1582.28. Marra CA, Esdaile JM, Guh D, Kopec JA, Brazier JE, Koehler BE,Chalmers A, Anis AH: A comparison of four indirect methodsof assessing utility values in rheumatoid arthritis.  Med Care2004, 42:1125-1131.29. Streiner DL, Norman GR: Health Measurement Scales: A Prac-tical Guide to Their Development and Use.  Oxford: OxfordUniversity Press; 1989. 30. Patrick DL, Erickson P: Health Status and Health Policy: Qualityof Life in Health Care Evaluation and Resource Allocation.31. Sinclair AJ, Burton JFJ: Development of a schedule for compen-sation of non-economic loss: quality of life values vs. clinicalimpairment rating.  In Research in Canadian Workers' CompensationEdited by: Chaykowski RP, Thomason T. Kingston, Ontario: IndustrialRelations Centre, Queen's University Press; 1995:123-140. 32. Shrout PE, Fleiss JL: Intraclass Correlations: Uses in assessingrater reliability.  Psychol Bull 1979, 2:420-428.33. Guyatt GH, Berman LB, Townsend M, Pugsley SO, Chambers LW: Ameasure of quality of life in clinical trials in chronic lung dis-ease.  Thorax 1987, 42:773-778.34. Nichol G, Llewellyn-Thomas HA, Thiel EC, Naylor CD: The rela-tionship between cardiac functional capacity and patients'symptom-specific utilities for angina.  Med Decis Making 1996,16:78-85.35. Albertsen PC, Nease RF, Potosky AL: Assessment of patient pref-erences among men with prostate cancer.  J Urol 1998, 159:158-163.36. Howard GS, Ralph KM, Gulanick NA, Maxwell SE, Nance D, GerberSL: Internal invalidity in pretest-posttest self-report evalua-tions and a reevaluation of retrospective pretests.  Appl PsychMeas 1979, 3:1-23.37. Daltroy LH, Larson MG, Eaton HM, Phillips CB, Liang MH: Discrep-ancies between self-reported and observed physical functionin the elderly: the influence of response shift and other fac-tors.  Soc Sci Med 1999, 48:1549-1561.38. Howard GS, Schmeck RR, Bray JH: Internal invalidity in studiesemploying self-report instruments. A suggested remedy.  JEdu Meas 1979, 16:129-135.39. Golembiewski RT, Billingsley K, Yeager S: Measuring change andpersistence in human affairs: types of change generated byOLD designs.  J Appl Behav Sci 1976, 12:133-157.40. Sprangers MAG, Schwartz CE: Integrating response shift intohealth-related quality-of-life research: a theoretical model.Soc Sci Med 1999, 48:1507-1515.41. Kaplan RM, Coons SJ: Relative importance of dimensions in theassessment of health-related quality of life for patients withhypertension.  Prog Cardiovasc Nurs 1992, 7:29-36.42. O'Boyle CA, McGee H, Hickey A, O'Malley K, Joyce CR: Individualquality of life in patients undergoing hip replacement.  Lancet1992, 339:1088-1091.43. Gorbatenko-Roth KG, Levin IP, Altmaier EM, Doebbeling BN: Accu-racy of health-related quality of life assessment: What is thebenefit of incorporating patients' preferences for domainfunctioning?  Health Psychol 2001, 20:136-40.44. Feeny D, Torrance GW: Incorporating utility-based quality-of-life assessments in clinical trials: Two examples.  Med Care1989:190-204.45. Torrance GW, Furlong W, Feeny D: Health utility estimation.Expert Rev Pharmacoeconomics Outcomes Res 2002, 2:99-108.46. Feeny D: A utility approach to assessing health-related qualityof life.  Med Care 2000, 38:S151-S154.47. Dyer J, Sarin R: Relative risk aversion.  Mgmt Sci 1982, 28:875-886.48. Bosch JL, Hunink MG: The Relationship between descriptiveand valuational quality-of-life measures in patients withintermittent claudication.  Med Decis Making 1996, 16:217-225.49. Bleichrodt H, Johannesson M: An experimental test of a theoret-ical foundation for rating scale valuations.  Med Decis Making1997, 17:208-216.50. Parducci A, Wedell D: The category effect with rating scales:number of categories, number of stimuli, and method ofpresentation.  J Exp Psychol 1986, 12:496-512.51. Torrance GW: Social preferences for health states: an empir-ical evaluation of three measurement techniques.  Socio EconPlan Sci 1976, 10:129-136.52. Wolfson AD, Sinclair AJ, Bombardier C, McGeer A: Preferencemeasurements for functional status in stroke patients: inter-rater and inter-technique comparisons.  In Values and LongTerm Care Edited by: Kane R. Lexington, MA: D.C. Heath;1982:191-214. 53. Feeny D, Townsend M, Furlong W, Tomkins DJ, Robinson GE, Tor-rance GW, Mohide PT, Wang Q: Assessing Health- RelatedQuality-of-Life in Prenatal Diagnosis, Comparing ChorionicVilli Sampling and Anmiocentesis: A Technical Report.Hamilton, Ontario: Centre for Health Economics and Policy Analysis,Page 9 of 10(page number not for citation purposes)New York, NY: Oxford University Press; 1993. McMaster University; 2000.  Working Paper 00-04Publish with BioMed Central   and  every scientist can read your work free of charge"BioMed Central will be the most significant development for disseminating the results of biomedical research in our lifetime."Sir Paul Nurse, Cancer Research UKYour research papers will be:available free of charge to the entire biomedical communitypeer reviewed and published immediately upon acceptancecited in PubMed and archived on PubMed Central Health and Quality of Life Outcomes 2006, 4:25 http://www.hqlo.com/content/4/1/2554. Krabbe PFM, Essink-Bot ML, Bonsel GJ: The comparability andreliability of five health-state valuation methods.  Soc Sci Med1997, 45:1641-1652.55. Furlong W, Feeny D, Torrance GW, Goldsmith CH, DePauw S, ZhuZ, Denton M, Boyle M: Multiplicative Multi-Attribute UtilityFunction for the Health Utilities Index Mark 3 (HUI3) Sys-tem: A Technical Report.  Hamilton, Ontario: Centre for HealthEconomics and Policy Analysis, McMaster University..  Working Paper98-1156. Le Galès C, Buron C, Costet N, Rosman S, Slama G: Développe-ment d'un index d'etats de santé pondéré par les utilités enpopulation française: le Health Utilities Index.  Economie etPrévision 2001, 150-1:71-78.yours — you keep the copyrightSubmit your manuscript here:http://www.biomedcentral.com/info/publishing_adv.aspBioMedcentralPage 10 of 10(page number not for citation purposes)


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items