Measuring Preferences for Cost-Utility Analysis: How Choice of Method May Influence Decision-Making

Abstract

Preferences for health are required when the economic value of health care interventions are assessed within the framework of cost-utility analysis. The objective of this paper is to review alternative methods for preference measurement and to evaluate the extent to which method may affect health care decision making. Two broad approaches to preference measurement that provide societal health state values are considered: 1) direct measurement and 2) preference-based health state classification systems. Empirical evidence comparing systems and decision making impact suggests that preferences will have the greatest impact on economic analyses when chronic conditions or long-term sequelae are involved. At present, there is no clearly superior method and further study of cost-effectiveness ratios from alternative systems is needed to evaluate system performance. Among studies that compared alternative preference-based systems, EQ-5D tended to provide larger change scores and more favorable cost-effectiveness ratios than HUI2 and 3, while SF-6D provided smaller change scores and less favorable ratios than the other systems. However, these patterns may not hold for all applications. Although there is some evidence that incremental cost-effectiveness ratio (ICER) thresholds (e.g., $25,000 per QALY gained) are used in decision making, they are not strictly applied and are complemented by consideration of legal, regulatory, and social equity issues. Nonetheless, as ICERs rise, the probability of acceptance of a new therapy is likely to decrease, making the differences in QALYs obtained using alternative methods potentially meaningful. It is imperative that those conducting cost-utility analyses characterize the impact that uncertainty in health state values has on the economic value of the interventions studied. Consistent reporting of such analyses would provide further insight into the policy implications of preference measurement.

Introduction

As pressures to contain costs of medical care have escalated, cost-utility analysis (CUA) has received both critical acclaim and scrutiny as a methodology to inform decision makers regarding the economic value of health care interventions.[1-6] The number of published CUAs has steadily grown over the past 20 years,[3] and use by policy-makers appears to be increasing.[7, 8] In some jurisdictions, including the UK, Australia, and the Canadian provinces of Ontario and British Columbia, a formal role for CUA in pharmaceutical coverage decisions is mandated.[7, 9] In other jurisdictions, such as the US, explicit use of CUA is more limited.[1, 10-13]

A key source of initial resistance to CUA was concern about the validity and comparability of results between studies.[1, 7, 14, 15] Issues of comparability have the potential to undermine a fundamental strength of CUA, which is meant to facilitate valid economic comparisons across a wide spectrum of interventions. Studies highlighting discrepancies between methods[15-22] underscore the importance of understanding the potential impact that methodological differences may have on decision making.

Development of a Reference Case by the U.S. Panel on Cost-Effectiveness in Health and Medicine[15] provided one methodological standard for CUA. The perspective recommended for the Reference Case is societal, and the methods for valuing health outcomes include use of a generic health state classification system and community preferences, with sensitivity analysis to include patient preferences for studies of specific conditions.[15] Since publication of the Reference Case, there is evidence of improving quality in CUA methodology and reporting.[3] Support for use of reference case criteria to improve comparability is indicated by recent recommendations within the US regulatory environment.[23] Formal guidelines in other jurisdictions demonstrate that although differences exist, there may be emerging consensus on key points noted in the Reference Case[24, 25] Despite these positive developments, the extent to which methodological differences in CUA may affect policy decisions remains uncertain. In this paper, we explore how alternative methods that fulfill Reference Case criteria for preference measurement, thereby yielding “societal health state values”, may affect decision making. We describe potential sources of variation in societal health state values, highlight relevant studies, and discuss research assessing decision making impact.

Methods for Estimating Societal Health State Values

The quality-adjusted life year (QALY) is the most commonly used measure for health in CUA.[26] QALYs combine the attributes of length and quality of life into a single measure. The length of time in each health state is weighted according to an associated “health state value”, on a scale anchored at 1 representing best imaginable health and 0 representing death. In our discussion, we use “preference” as a general term reflecting a health state's desirability and “health state value” (HSV) to connote the numerical strength of preference for a health state.

We consider two general approaches to HSV measurement: 1) direct preference elicitation for relevant health states and 2) preference-based health state classification systems.[27-29]

Direct Preference Measurement

The standard gamble, time tradeoff, and rating scale are commonly used preference elicitation methods with unique characteristics.[15, 30-32] Differences between preferences obtained by these methods are well-documented,[21, 33-35] with HSVs typically highest for standard gamble and lowest for rating scale.[36-38] A recent study comparing elicitation methods for EQ-5D and SF-6D concluded that this component alone can contribute to differences in HSV up to 0.31, and may impact on cost-effectiveness ratios.[39] One study of dialysis patients provided evidence that patient preferences from the standard gamble resulted in cost-effectiveness ratios that were higher by $5,916 per QALY than when time tradeoff values were used. Although this is one example of how measurement approach may affect cost-effectiveness results, individual patient rather than societal HSVs were used.[21]

Direct measurement of societal HSVs is a resource intensive endeavor, requiring development of relevant health state descriptions, and access to a representative population sample. Since this is often not feasible, researchers may use condition, age and sex-specific values from a published source[40]

Preference-Based Health State Classification Systems

Preference-based health state classification systems define each respondent's health state based on a questionnaire and assign a societal HSV with a scoring algorithm that incorporates preferences from a general population sample. This approach allows researchers to use societal HSVs with minimal resources compared to direct preference measurement.

The most widely used systems[19] include the EuroQol 5D (EQ-5D),[41-43] the Health Utilities Index (HUI),[44, 45] the Quality of Well-Being Scale (QWB),[46] and the SF-36-derived SF-6D.[47, 48] Three basic steps in developing a system are 1) classifying health, 2) eliciting population preferences for a subset of health states, and 3) developing a scoring algorithm to assign values for the full range of health states. We address potential sources of variation and related research for each step.

Descriptive System

A generic self-report health status questionnaire is the basis for most systems. Most systems include attributes for pain, physical function, social or role function, and anxiety/depression, but differ in the number of response levels and how these are described and weighted. Some systems include other attributes, such as hearing and vision. Other differences include the perspective used in assessing health status. For example, HUI questions ask about respondents' functional capacity,[44, 49, 50] while EQ-5D and others ask about actual performance. The reference period also varies among systems, from “today” to “over the past four weeks”.

Each instrument characterizes a unique number of health states based on the numbers and levels of attributes included in the questionnaire. It is unknown how many health states are needed to describe health adequately, however, EQ-5D has 243 health states, HUI2 has 24,000 and HUI3 has 972,000.

Taken together, these aspects of the descriptive systems would have an impact on psychometric properties of the system, including the ability to measure health status numerically, and to detect meaningful changes in health. Inadequacy in the descriptive systems may result in ceiling and floor effects, inability to measure key attributes of health (validity), and inability to measure important change (responsiveness).

Preference Measurement

In the development of preference-based systems, a subset of the unique system-defined health states are valued by a sample of the population as the basis for estimating values for the full range of health states. Choice of direct preference elicitation method implicitly incorporates differences noted earlier into the estimation of each system's societal HSVs. EQ-5D offers value sets based on both time tradeoff and rating scale measurements.[51] The SF-6D employs the standard gamble[47, 48] and the QWB uses category scaling[46, 52] methods. The developers of HUI and others, for example, argued that the rating scale is not appropriate for use in CUA. They used the standard gamble and a transformation of rating scale preference measurements for HUI2 and HUI3.[36] Debate continues about the merits of various elicitation methods[53] and power transformation.[54, 55]

Comparing CUA results using alternative preference measurement methods for the same system can demonstrate the isolated impact of preference elicitation method. Conner-Spady[56] compared HSVs from EQ-5D using the time tradeoff and visual analogue scale (VAS) in 436 joint replacement patients, and found a lower baseline mean, wider range, and more negative HSVs, and larger QALY gains for time tradeoff than VAS-based value sets, indicating a more favorable cost-effectiveness ratio for EQ-5D when using TTO relative to VAS. For example, the QALY gain reported for EQ-5D with TTO-based preference weights was 5.14, compared to 3.64 for the VAS using a 10-year time horizon. If the cost of joint replacement were $7,000, [57] the ICER would vary only slightly across methods (from $1,362 to $1,923 per QALY gained ) and be unlikely to influence decision making.

Source of Community Preferences

The representativeness of population samples varies among systems.[58] Furthermore, eliciting and scoring population preferences for health state classification systems are resource-intensive, and HSVs are not available for every population. Therefore, preferences from a different population than that of interest are sometimes used. The extent to which differences in population preferences would contribute to variation in CUA results has been explored. In a study comparing rating scale valuations of EQ-5D health states by Finnish and US general population samples, small differences were noted that the authors concluded would not impact EQ-5D HSVs in international studies.[59] A study comparing TTO HSVs between general US and UK population samples[60] reported higher HSVs for the US sample. Differences in EQ-5D TTO HSVs between UK and Spanish populations,[61] suggest that cultural differences may influence health state valuation.

Scoring Methods

Using directly measured preferences from a sample of the general population, a statistical model is fit to estimate HSVs for the remaining health states. The HUI scoring system is based on multiattribute utility theory[62] and a multiplicative function that captures interactions among attributes and allows characterization of single attribute utility functions for levels within each attribute.[36, 63] EQ-5D with York preference weights uses an additive model that includes level of severity, movement away from perfect health, and a term (the N3 term) to account for interaction between attributes when any attribute is at the worst level.[42, 43] Scoring algorithms have been developed for EQ-5D for various populations. U.S. TTO-based preference weights are now available for the EQ-5D (EQ-5D-US) using a random effects model.[64] SF-6D scoring is very similar to that of EQ-5D N3 model, with a smaller decrement when any attribute is at its worst level.[65] QWB utilizes an additive function that does not allow for interaction among the attributes.[46] The range of each scale and whether they include states worse than death (e.g. negative scores) vary.

Few studies compared CUA results using different scoring algorithms for the same system. Conner-Spady[56] assessed the effect of the N3 term on EQ-5D HSVs, and reported that it resulted in lower baseline mean and effect size for values, and more QALYs gained in a sample of joint replacement patients.

Empirical Evidence

We searched the international published literature (See Appendix for search terms), and found no studies that directly investigated impact of preference measurement method on policy decisions. Though a systematic review of policy decisions was beyond the scope of this paper, we included all identified studies that provided head-to-head comparisons of the most commonly used preference-based system or addressed impact of system choice.

Comparisons of Preference-Based Health State Classification Systems

A review of 23 published cost-utility analyses conducted alongside clinical trials[19] found that 20 studies utilized a preference-based health state classification system to estimate QALYs. HUI and EQ-5D were the most commonly used systems, with 16 using EQ-5D. The authors suggested that different systems could qualitatively impact CUA results, and called for greater reporting transparency.

Table I summarizes cross-sectional comparisons between systems. These studies found varying differences in mean HSVs across systems and offer insight into system characteristics that may contribute to variation in HSVs.[40, 66-83] Generally, correlations between values from alternative systems were moderate to strong, indicating that they measure the same construct. SF-6D demonstrated floor effects, and a limited range of available scores. EQ-5D did not provide HSVs between .88 and 1, and provided lower HSVs for similar health states from other systems. HUI3 was limited in characterizing diminished mobility other than ambulation. These characteristics may be important for instrument choice relative to the condition of interest.

Table I. Comparison of mean health state values in cross-sectional comparisons of systems.

Study (first author, year, n)	Systems Studied	Baseline Mean (SD)	Correlation Coefficient
			EQ5D	HUI2	HUI3	SF6D	15De
Luo, 2005, n = 4048	EQ-5D	0.87(0.13)		0.67a	0.66a
General adult US population	HUI2	0.86(0.32)			0.87a
	HUI3	0.81(0.38)
Petrou, 2005, n = 14,736	EQ-5D	0.84(0.23)				0.70a
General adult UK population	SF-6D	0.80(0.15)
Barton, 2005, n = 915	EQ-5D	0.79(0.23)			0.44b	0.50b
Hearing impairment	HUI3	0.56(0.15)				0.41b
	SF-6D	0.77(0.08)
McDonough, 2005, n = 2,097	EQ-5D	0.39(0.33)		0.68c	0.67c	0.69c
Spine disorders	HUI2	0.59(0.22)			0.78c	0.67c
	HUI3	0.45(0.27)				0.72c
	SF-6D	0.57(0.12)
Espallargues, 2005, n = 209	EQ-5D	0.72(0.22)	-	-	-	-	-
Macular degeneration	HUI3	0.34(0.28)
	SF-6D	0.66(0.14)
Marra, 2004, n = 313	EQ-5D	0.66(0.13)	-	-	-	-	-
Rheumatoid arthritis	HUI2	0.71(0.19)
	HUI3	0.53(0.29)
	SF-6D	0.63(0.24)
Feeny, 2004, n = 264	HUI2	0.92(0.13)	-	-	-	-	-
Teen survivors low birthweight	HUI3	0.84(0.16)
Maddigan, 2003, n=372	HUI2	0.78(0.17)	-	-	-	-	-
Type II diabetes	HUI3	0.64(0.29)
Luo, 2003, n = 114	EQ-5D	0.75(0.21)			0.45c
Rheumatic disease	HUI3	0.76(0.17)
O'Brien, 2003, n = 246	HUI3	0.61(0.12)				0.58a
Cardiac disease	SF-6D	0.58(0.16)
Schulz, 2002, n = 29	EQ-5D	0.81(0.12)			0.32c
Benign prostatic hyperplasia	HUI3	0.79(0.12)
Brazier, 2001d, n = 2605	EQ-5D	0.59(-)				0.66a
Combined patient groups	SF-6D	0.63(-)
Stavem, 2001, n = 348	EQ-5D	0.81(0.23)					0.78c
Epilepsy	15De	0.88(0.12)
Belanger, 2000, n = 1,477	EQ-5D	0.83(-)			0.69a
Canadian general population	HUI3	0.85(-)

^a= Pearson's correlation coefficient

^b= Kendall's tau for n = 863

^c= Spearman's rank correlation coefficient

^d= This study reported values for patient groups: COPD, osteoarthritis, irritable bowel syndrome, low back pain, leg ulcers, menopausal symptoms, and age greater than 65 years.

^e= 15D is a preference-based health state classification system for which data were reported compared to more commonly used systems [81]

^-= not reported

Evidence of differences in HSVs from cross-sectional studies is reason for concern, but longitudinal studies are necessary to understand system performance when measuring change in health. Fewer studies[56, 76, 84-98] report longitudinal head-to-head comparisons of preference-based systems (Table II). EQ-5D estimates were generally largest, followed by HUI3, HUI2, and finally SF-6D. Interactions have been noted between HSVs and level of improvement.[88] Pickard et al reported that change in SF-6D HSVs correlated with mental status, while change in HUI and EQ-5D related more to daily function and disability.[84] No consistent pattern of correlation was evident in the studies we reviewed.[84, 92]

Table II. Studies addressing longitudinal comparisons in mean health state values provided by the most commonly used preference-based health state classification systems.

Study (first author, year, n)	Systems Studied	Baseline Mean (SD)	Change in Health State Value Mean (SD)
Langfitt, 2006, n=64	EQ-5D	0.76(0.26)	0.11(0.28)
Chronic Epilepsy	EQ-5D-US	0.82(0.18)	0.07(0.21)
	HUI2	0.78(0.18)	0.03(0.15)
	HUI3	0.61(0.30)	0.08(0.27)
	SF-6D	0.70(0.14)	0.08(0.16)
Kaplan, 2005, n = 628	EQ-5D	0.56(0.25)	0.10(0.27)
Rheumatoid Arthritis	HUI2	0.64(0.20)	0.06(0.18)
	HUI3	0.43(0.27)	0.09(0.24)
	SF-6D(VAS)	0.43(0.16)	0.06(0.16)
	SF-6D(SG)	0.81(0.10)	0.03(0.09)
Pickard, 2005, n = 98	EQ-5D	0.31(0.38)	0.32(0.38)
Stroke	HUI2	0.51(0.20)	0.12(0.23)
	HUI3	0.19(0.30)	0.25(0.32)
	SF-6D	0.55(0.09)	0.13(0.15)
Stavem, 2005, n = 60	EQ-5D	0.77(0.26)	0.02(-)
HIV/AIDS	SF-6D	0.73(0.17)	0.03(-)
	15Da	0.86(0.14)	0.02(-)
Thoma, 2005, n = 41	HUI2	(-)	0.06(0.14)
Breast Reduction Surgery	HUI3	(-)	0.12(0.19
Hatoum, 2004, n = 184	HUI3	0.63(0.29)	0.15(0.28)
Coronary Artery Disease	SF-6D	0.67(0.12)	0.08(0.13)
Feeny, 2004, n = 63	HUI2	0.62(0.19)	0.22(0.07)
Total hip arthroplasty	HUI3	0.62(0.32)	0.23(0.04)
	SF-6D	0.61(0.10)	0.10(0.02)
Holland, 2004, n = 123	EQ-5D	0.61(0.29)	-0.16(0.34)
Elderly patients	AQoLb	0.45(0.27)	-0.12(0.24)
Longworth, 2003, n=183	EQ-5D	0.52(0.33)	0.09(0.37)
Liver transplant	SF-6D	0.61(0.12)	0.01(0.28)
Bosch, 2002, n = 87	HUI2	0.70(0.20)	0.10(-)
Intermittent claudication	HUI3	0.66(0.20)	0.11(-)
	SF-6D	0.66(0.09)	0.08(-)
Bosch, 2000, n = 88	EQ-5D	0.57(0.25)	0.22(-)
Intermittent claudication	HUI3	0.66(0.20)	0.11(-)
Suarez-Almazor, 2000, n = 37	EQ-5D	0.38(0.33)	-4.8 (17.4)c
Low back pain	HUI2	0.49(0.19)	-1.8 (16.1)c

^a15D is a preference-based health state classification system for which data were reported compared to most commonly used systems [81]

^bAQoL is a preference-based health state classification system for which data were compared to most commonly used systems [76, 95]

^c= 3 month results, rescaled to 1-100, with 100 corresponding to best health

^-= Not reported

Comparisons of ICERs obtained using alternative systems were less common. Thomas et al[99] reported CUA results for chronic low back pain acupuncture treatment using SF-6D and EQ-5D. The ICER for SF-6D was £4,241(95%CI 191-28,026) compared to £3,598 (95%CI 189-22,035) for EQ-5D. Neumann et al[94] compared the results of a cost-effectiveness model for Alzheimer's drug treatment using HUI2 and HUI3, and found lower mean utility scores using HUI3 versus HUI2, resulting in ICERs of $9,000/QALY for HUI3 and $11,000/QALY for HUI2 with duration of drug effect of 18 months. The authors noted that while the difference in ICERs was slight for this analysis of drug treatment, it could be substantial for a disease prevention drug..

Overall, EQ-5D tended to have larger changes in HSVs, which generally translate to more favorable cost-effectiveness ratios when using EQ-5D compared to HUI2 or HUI3. Similarly, in studies that included EQ-5D and SF-6D, EQ-5D had larger changes. Comparing HUI2 or HUI3 and SF-6D, SF-6D tended to have smaller changes. These patterns are generally consistent with the results of cross-sectional comparisons, yet it is difficult to identify a superior system.

Potential Impact of System Choice on Decision Making

Recent commentaries illustrate the concern that both lack of data on preferences and variation in methods will impact health care decisions[100, 101]. We found no empiric studies that address this question by directly investigating policy decisions relative to different measurement methods. Studies addressing the potential influence of economic evaluation on policy decisions have considered thresholds for favorable cost-effectiveness to characterize qualitative changes in study results. Chapman et al[17] assessed the affect of quality adjustment on analyses reporting both life years and QALYs published before 1998 and found qualitatively similar results for both QALYs and life years for the majority of analyses. However, quality adjustment had an important impact, causing estimates to cross thresholds for dominance of $50,000, or $100,000, in 18% of cases. This could be interpreted as evidence that in the majority of cases, method choice would not be likely to have a qualitative impact on the results of the studies, and therefore on downstream policy decisions. Use of QALYs had the greatest impact for analyses assessing palliative treatments, chronic diseases, and situations where there may be negative long-term side-effects. This may indicate that for this subset of diseases, differences in estimated HSVs carried over a lifetime may have a qualitative impact on economic evaluation results and related policy decisions.

Using the same database, Schackman et al[18] examined the results of sensitivity analyses on health-related quality of life for pharmaceutical CUAs to determine the proportion of times that specific cost-effectiveness thresholds (i.e., $20,000, $50,000 and $100,000 per QALY gained) were crossed. In 31% of sensitivity analyses, the ICER exceeded a threshold and in 13% of analyses, the ratio fell below a threshold. This indicates that if the magnitude of preference method-related variation reached the levels modeled in sensitivity analyses, there would be qualitative differences in results, as defined by crossing thresholds. The use of thresholds for policy decisions among a broad range of organizations is not well documented. We found some indication that thresholds of £20-30,000 are loosely used, and that overall, as ICERs increase, the likelihood of acceptance probably decreases.[102-105]

Discussion

The paucity of published evidence limits our ability to draw conclusions concerning how preference measurement method affects policy decisions. The ICERs available for comparison do not represent large enough changes to impact decision making. However, evidence from comparisons of preference-based systems support a wide range of variation in estimates. Based on the evidence reviewed here, it appears that choice of preference measure may contribute to qualitatively different ICERs under some circumstances. As ICERs rise, the probability of acceptance appears to decrease, making the differences in QALYs obtained using alternative methods potentially meaningful. This is especially important for treatments with long-term consequences and ICERs around common thresholds. In our review, EQ-5D tended to provide more favorable cost-effectiveness ratios than HUI, while SF-6D provided less favorable ratios than the other systems. Whether these patterns will hold for all applications depends on each system's ability to measure change across the full range of health.

To assess the impact of health state valuation on cost-utility analysis, our review focused on societal HSVs estimated with the most commonly used preference-based health state classification systems. However, evidence shows that in practice, the majority of CUAs do not yet meet Reference Case criteria[20, 106, 107]. Reviews indicate that use of preference-based systems appears to be increasing, with approximately 23% of published CUAs using community preferences, and 20% estimated from preference-based systems.[20, 107] Direct valuations using standard gamble, time tradeoff, and rating scale were used 35.3% of the time, with 26.8% of valuations from patients.

Arguments for using community preferences hold that the population potentially affected by the decision should be polled. From a position of uncertainty about their own future health (i.e., ‘the veil of ignorance’) they would value policy decisions with the most benefit for society as a whole.[15, 108] However, patients or population subgroups may be in a better position to assess the impact of condition-related health changes than a general population sample.[15, 109] This is controversial in light of evidence that patients provide higher values than other groups for their health states,[15, 33, 67, 68, 75, 108, 110-112] and the possibility that this could cause undervaluation of preventive intervention[67] and treatment. In addition, there is evidence that differences in valuations between patients and other groups varies with other factors,[113, 114] such as severity and chronicity of illness.[115, 116]

As a practical matter, the choice of methods may be dictated by resource availability, making off-the-shelf tools attractive. To facilitate cost-utility analyses when primary data are not available, the Panel on Cost-Effectiveness in Health and Medicine recommended construction of a national catalogue of “off-the-shelf” community preferences for health states.[15] Work is ongoing in this area and includes population-based HSVs using multiple instruments,[40, 117-120] mapping of health status data to HSVs, and catalogues or registries from the published literature.[20, 121] Several methods have been explored to estimate HSVs from the SF-36,[66, 122-125] the SF-12,[122, 126-135] condition-specific tools,[136-138] and national survey instruments.[139-141] Some studies indicate that estimates from mapping may have important limitations[92, 122, 142] while others emphasize the fairly strong correlations between approaches.[23] A study investigating potential decision making impact of algorithm choice compared ICERs from various SF-36 and SF-12 algorithms and the SF-6D.[143] They reported ICERs for an asthma cohort ranging from $30, 769 to $63, 492 per QALY, and for a stroke cohort of $27, 972 to $50,000 per QALY. Many of the reported ICERs did not have overlapping confidence intervals. The growth of methods for estimating preferences using existing data is expanding the range of possible CUA applications, and warrants continued comparative study. In addition, efforts have been made to make published cost-utility analysis more accessible through registries,[144-147]some of which include information on methodology. These efforts are important steps in improving availability and transparency of cost-utility analyses.

In conclusion, it is unknown to what extent choice of preference measurement method impacts health policy decision-making. The existing evidence points toward potential impact for a subset of situations, most likely for ICERs near established thresholds when chronic diseases and/ or long-term health effects are involved. At present, there is no clearly superior method for estimating societal preferences, however, alternative systems produce a wide range in HSVs. The ability to convert health state values from one system to another would greatly improve comparability of CUAs. A recent publication of ongoing research using MEPS and National Health Interview Survey data uses linear regression to standardize estimates from various systems, and to convert estimates from one system to those of another.[148] Related research provided population-based norms.[149] Ongoing work will further our understanding of system responsiveness.[23]

Psychometric comparisons between systems were more common than comparisons of impact on ICERs and/or policy decisions. Explicit study of cost-effectiveness ratios obtained from alternative preference-based systems is needed to improve our understanding of potential policy implications. We encourage researchers who have access to comparative data to publish these estimates. Sensitivity analyses are suggested to address uncertainty associated with estimates of health effect, including variation due to measurement method. An in-depth systematic review of decisions made by policy makers relative to alternative measurement method would be a timely and important contribution to the literature. Research in these areas combined with ongoing dialogue about standardization and conversion of estimates hold promise for attaining consistency of estimates for CUA.

Appendix

The following key words were used in Medline to identify papers addressing the topic of preference measurement method and policy decision-making. In addition, selected author, references from papers known from previous work, and known web sites were hand-searched. Searches were limited to English language.

• Cost-utility analysis

• Cost-effectiveness analysis

• Economic evaluation

• Methods

• Attitude to Health

• Health Status Indicators

• Health Status

• Quality of Life

• Quality-Adjusted Life Years

• QALYs

• Time tradeoff

• Standard gamble.

• Rating scale

• Visual analogue scale

• Utilities

• Health state values

• Values

• Preferences

• Preferences for health states

• Preference classification systems

• Preference-based

• EQ-5D.

• HUI

• SF-6D

• SF

• Quality of Well-Being.