DETERMINING THE SOCIETAL VALUE OF A QALY BY SURVEYING THE PUBLIC IN ENGLAND AND WALES: A RESEARCH PROTOCOL SUBMISSION TO NATIONAL INSTITUTE FOR CLINICAL EXCELLENCE AND NATIONAL CO-ORDINATING CENTRE FOR RESEARCH METHODOLOGY
Rachel Baker1 Sue Chilton2 Cam Donaldson1,2 Michael Jones-Lee2 Hugh Metcalf2 Phil Shackley1 1
School of Population & Health Sciences and 2The Business School University of Newcastle upon Tyne Mandy Ryan Health Economics Research Unit University of Aberdeen
Name and address for correspondence: Professor Cam Donaldson University of Newcastle upon Tyne School of Population and Health Sciences Centre for Health Services Research 21 Claremont Place Newcastle upon Tyne NE2 4HH Tel: 0191 222 3463 Fax: 0191 222 6043 e-mail:
[email protected]
CONTENTS Page Executive summary
i
1. Introduction
1
2. Background: 2.1 WTP for a QALY: the health economics literature 2.1.1 Theoretical perspectives on WTP for a QALY 2.1.2 Empirical estimates of WTP for a QALY 2.1.3 Beyond QALYs: is there value in other characteristics of health care or its beneficiaries? 2.2 Lessons from safety and environment literatures 2.3 Summary of issues to be addressed 2.3.1 The basic approaches 2.3.2 Additional important issues
3 3
3. Estimating a WTP-based monetary value of a QALY by the contingent valuation method: an assessment of six possible approaches and a suggested way forward: 3.1 Simplifying assumptions 3.2 Six possible approaches 3.3 The way forward
10 14
16 16 16 24
4. Estimating a WTP-based value of a QALY by the use of DCEs: 4.1 Defining the attributes for the baseline valuation 4.2 Modelling a WTP-based value of a QALY
27 27 28
5. Going beyond the baseline estimates of the WTP-based values of a QALY 5.1 Testing for scope 5.2 Testing for informational (or “process”) effects 5.3 Testing for impact of health gains being in terms of quality of life only, age and health status on self-only valuations 5.4 Using the DCE approach to investigate broader societal concerns
30 30 30 31
6. Survey methods 6.1 Using the three approaches to derive a baseline value (Aim 1) 6.2 Comparisons of CV/SG chained, WTP-for-gains-in-life-expectancy and DCE-based values with existing values of safety (Aim 2) 6.3 Investigating other factors (Aim 3) 6.3.1 Testing for scope 6.3.2 Testing for informational (or “process”) effects 6.3.3 Testing for impact of health gains being in terms of quality of life only, age and health status on self-only valuations 6.4 Using the DCE approach to investigate broader societal concerns 6.5 Timetable and budget 6.5.1 Timetable 6.5.2 Budget
33 33 36
Appendix A: Willingness-to-pay-based values of safety in public appraisal
sector project
40
Appendix B: Summary of quality-of-life value used by NICE in health technology appraisals
46
References
58
31
36
37 38
EXECUTIVE SUMMARY This proposal provides an outline of the justification and methods for three different approaches to estimating a willingness-to-pay-based value of a quality adjusted life year (WTP-based value of a QALY). The momentum for this research stems from: •
• •
the evaluation of technologies undertaken by the National Institute for Clinical Excellence (NICE), which raises questions about the monetary value to attach to health gains (through enhanced survival and quality of life) produced by such technologies; consideration of whether aspects of health care beyond the health effects captured by QALYs can be captured by any other metric; and ultimately, how the benefits arising from investments in health care can be compared with those arising from investments in other areas of the public sector which are not (and often cannot be) valued in terms of QALYs.
In health economics, a theoretical strand of literature has attempted to link the concepts of WTP and QALYs. However, it is not known whether the postulated relationships hold in practice. Empirically, only indirect attempts have been made to link WTP and QALY estimates of gains produced by different procedures. Parallel developments in the valuation-of-safety literature have led to estimates of the value of (statistical) lives, which, despite being adapted to provide useful rough estimates of WTP-based values of a QALY, do not account for quality-of-life gains (or can do so only by assumption). Thus, there is a need for a monetary (WTP-based) value of a QALY in policy terms and existing literature reveals no studies which have directly estimated such a value. In this document, cases are made for three possible approaches to estimating a WTPbased value of a QALY. Given the challenges involved, it is recommended that all three approaches be tested, these being: •
•
•
a contingent valuation/standard gamble chained approach, whereby respondents would provide a monetary value for returning from an impaired condition (which would otherwise last for a year) to normal health and also a standard gamble question through which the same condition is valued relative to a year of healthy life, the combination of these allowing a WTP-based value of a QALY to be calculated; a WTP-for-increased-life-expectancy approach whereby respondents are asked to value increases in life expectancy of different durations in normal health, from which values for durations not asked about in the questionnaire can be modelled (with questions also being asked about equivalent QALY gains when made up of quality and length of life improvements); a DCE approach, whereby respondents are faced with scenarios involving different degrees of length of life, quality of life and forgone money, and are asked to choose between the scenarios, so permitting the researchers to observe how these attributes are traded off against each other and, thus, values
i
of length of life, quality of life, and the two combined, to be portrayed in terms of the money attribute. Use of all three of these methods will permit “triangulation” towards a basic WTPbased value of a QALY. Of course, even if each approach produced the same valuation, it would be doubtful if such a value would apply in all possible situations. Also, there are some methodological “acid tests” which any such method must pass for the results it produces to be considered plausible. Therefore, the second stage of the research will test how such baseline values vary according to characteristics of health care, such as: • • •
scope (i.e. the size of the health gain considered); information about the process as well as the outcome of care; whether QALY gains are made up of survival only or quality as well as length-of-life gains;
and broader societal concerns, such as: • • • •
patients’ initial severities of illness; age; type of disease; culpability.
Baseline WTP-based values of a QALY will be estimated through the conduct of 150 focus groups in four sites across England and Wales. These will involve 750 participants who will be asked to complete a questionnaire whilst in the focus group. Because of the different scenarios involved, 180 focus groups involving 930 participants will be conducted to test for the effects of other characteristics of health care and broader societal concerns. The implications of such work are profound, as WTP-based values of a QALY, if thought to be plausible, will likely be used as an important part of the decisionmaking process over what guidance to recommend to the NHS vis-à-vis the adoption of welfare-enhancing, but often net resource consuming, technologies.
ii
1.
INTRODUCTION
To date, the dominant evaluation paradigm in health economics has been cost-utility analysis, in which the benefits of health care are valued in terms of quality adjusted life years (QALYs). 1 This poses challenges when making decisions based on evaluations of single health-care interventions, which inevitably involve judgements about whether the QALYs gained are worthwhile, or, in other words, what is the monetary value of a QALY. This issue has come to the fore in UK health policy as a result of the creation of the National Institute for Clinical Excellence (NICE) (Devlin and Parkin, 2002; Loomes, 2002). In offering guidance to the National Health Service (NHS) about the uptake (or maintenance) of an intervention, NICE has to weigh up the costs and benefits involved. If it is thought necessary to have these costs and benefits expressed in a common metric, usually money, the question is raised as to what value to place on improvements in length and quality of life. This is similar to the need of the Department for Transport to place a value on human life saved by reductions in risk arising from road safety improvements, except that the challenge in health care is, perhaps, greater, due to the additional complication of valuing quality of life in conjunction with length of life. As explained below, the (theoretically) best approach to estimating the monetary value of such benefits is to use the concept of “willingness to pay”2. The research outlined in this protocol, therefore, will address the issue of establishing a WTP-based value of a QALY through several methodological routes. Thus, the main aims of the research outlined in the protocol are to: 1.
2.
obtain baseline empirical estimates of WTP-based values of a QALY, when QALY gains comprise (a) gains in life-expectancy alone and (b) life years adjusted for their quality. Theory and standard approaches treat the value of such QALY gains as being equal (see reference to Bleichrodt and Quiggin (1999) in sub-section 2.1 below), but will they be in practice? If not, what are the implications for application/policy? These empirical estimates would be generated by more than one procedure (for “triangulation” purposes). The procedures include the following, all of which are explained in detail below: • a contingent valuation/standard gamble (CV/SG) chained approach; • WTP for gains in life-expectancy in normal health (1 month, 3 months, 6 months); and • discrete choice experiments (DCEs). compare these empirical estimates with various back-of-the-envelope estimates derived from existing values for prevented fatalities (VPFs) (such as the Department-for-Transport figure of £1.25 million). These back-of-theenvelope estimates will be based on refinements of the first four approaches presented in section three below; note, for reasons outlined in the protocol, that these four approaches rely on the (arguably defensible) “groupaggregation” of individual WTP for marginal gains rather than the (arguably
1
Of relevance to studies referred to below, the cost-per-QALY approach is often referred to in the health economics literature as cost-effectiveness analysis (CEA). 2 Or willingness to accept compensation.
1
3. 4.
indefensible) linear-proportionately “scaling up” of individual WTP for marginal gains. examine the manner in which other factors (e.g. “process” of care, size of health gain, age and health status) might impact on the baseline values of a QALY. extend the analysis (which so far is on a self-only basis) to accommodate wider “societal” concerns, thus attempting to value factors such as age, severity of initial health state and culpability, with the respondent taking the perspective of a more detached “citizen”. This could be done using all three approaches listed in 1., but is probably best accommodated through the use of DCEs.
A project meeting these four aims will address the second and third research issues listed in the initial invitation to tender issued by NICE and the National Coordinating Centre for Research Methodology (NCCRM), these being “What value do people put on QALYs in terms of other ‘benefits’?” (in the case of this proposal, in cash) and “How does the value people put on a QALY vary?”. In the following section, to make the case for the proposed research, the current “state of the art” in estimating WTP for a QALY is assessed. This is done by reviewing two major areas of literature, covering previous theoretical and empirical work in the health field as well as from research on valuation of safety policies. Lessons learned from these fields then lead us to suggest important next steps for research in this field. In section three, various theoretical approaches to WTP-based valuation of a QALY, based on the general method of contingent valuation, are introduced, with cases made for further research on two of these. This is then followed, in section four, by an outline of a DCE-approach to estimating a WTP-based value of a QALY, with the case being made that this approach provides another method from which to triangulate a baseline monetary value. Section five outlines how each of the approaches used to estimate baseline WTP-based values of a QALY can be adapted to take account of other aspects of health care and broader societal concerns. The sixth section of the protocol outlines the survey methods to be used in administering the CV/SG chained, WTP for increased life expectancy and DCE approaches and, thus, in establishing initial values of a QALY, before describing further experiments aimed at testing the impacts of other factors, such as age, health status, aspects of health care and societal concerns. Qualitative methods are built in throughout by having focus groups as the main mode of collection for both quantitative and qualitative data (Chilton et al., 2002) and the use of follow-up interviews with respondents (Shiell et al., 2000). As the methods section progresses to address a more and more complex set of issues, the project becomes more resource intensive. The sixth section, therefore, also provides more administrative details, such as a timetable and proposed budgets for two different sizes of potential projects.
2
2.
BACKGROUND
NICE has a requirement “to produce a common currency of effectiveness for the NHS”. Of course, it may be possible to compare competing interventions on the basis of cost per QALY gained. However, given that many of the recommendations made by NICE involve considerations of the costs and benefits of single interventions, the question of the value of QALY gains is raised. This is further exacerbated by the challenges of considering aspects beyond the health effects captured by QALYs and how benefits produced by health care can be compared with those arising from investments in other areas of the public sector which are not (and often cannot be) valued in terms of QALYs. Hence the need for a monetary measure to reflect the value of a QALY. The monetary measure which we propose is based on the method of “willingness to pay” (WTP). In standard welfare economics, maximum WTP represents the theoretically-correct measure of “strength of preference” for, or value of, a commodity (Mishan, 1971; Pauly, 1995). In areas of public sector activity, such as health care, in which conventional markets do not exist, decisions still have to be made about how best to use limited resources. This requires valuation of both resource costs of interventions and their benefits (the benefits being health gain and other sources of well-being), elicited in surveys by use of hypothetical WTP questions. This is essentially the contingent valuation approach in which respondents state values for the good in question rather than reveal them as a result of real market-based choices, the reasons for this being covered in more depth in sub-section 2.2. WTP focuses on the valuation of benefits, whereby a health care option may be described to a respondent and the person is asked what is his/her maximum WTP for it. In principle, with this type of information, the combination of NHS interventions could be chosen which maximises the value of benefits to the community. It is important to distinguish WTP, as a measure of benefit, from the cost of provision of a good. For any good, many people would be willing to pay more than its cost. Given that the good is provided at cost, and many would be willing to pay more, it is the maximum WTP for the good that represents its benefit to these individuals. For any individual, the difference between this benefit and the cost of the good represents a gain in well-being from having the good provided. 2.1
WTP for a QALY: the health economics literature
The WTP method was first applied in the health area in the famous study of WTP to avoid heart attacks (Acton, 1976). Subsequent to that, there were relatively few studies in the area of health (Diener et al., 1998), probably as a result of the QALY being perceived as a more acceptable measure of benefit than one which valued life in monetary terms. 3 It was not until the publication of two empirical papers in the Journal of Health Economics in the early 1990s (Donaldson, 1990; Johannesson et al., 3
On the face of it, it would seem that it is problematic to use WTP measures to inform decisions about the allocation of resources for commodities, such as health care, for which such allocation is supposed to be on the basis of (some notion of) need. This is because WTP is obviously associated with ability pay. However, it has been shown that this need not impede the use of WTP in health care economic evaluations (Donaldson, 1999) and also that whatever method is use to value benefits, including QALYs, it will suffer from the same distributional concerns (Donaldson et al., 2002).
3
1991) and the conceptual paper by Gafni (1991) that the feasibility of using the WTP approach in health economics was once again recognised and more studies began to be undertaken (Diener et al., 1998; Klose, 1999; Olsen and Smith, 2001). In fact, Gafni was one of the few people to undertake empirical work in the health field during the 1980s, with his small study of the WTP of women on a kibbutz for contraception services (Gafni and Feder, 1987). Although QALYs may appear to have been introduced to health economic evaluation after Acton’s initial WTP study (Torrance, 1984; Williams, 1985), they have emerged as the dominant valuation paradigm in this field, partly for the reasons outlined in footnote 3 and expanded upon by Donaldson and Shackley (2003). More recently, two strands of literature, one theoretical and one empirical, have strived to link the two approaches. However, as will be seen, the empirical literature is not strong, thus reinforcing the case for the research proposed in this protocol. 2.1.1 Theoretical perspectives on WTP for a QALY Within health economics, the need for a WTP-based value for a QALY stems from the largely theoretical debate in the literature about the equivalence (or nonequivalence) of cost-benefit analysis (CBA) and CEA: “Cost-effectiveness analysis should then be interpreted as an estimation of the cost function to produce health effects….In order to decide whether a treatment is cost-effective or not the estimation of the cost function has to be supplemented with information about the willingness to pay per unit of health effects to decide whether benefits exceed costs or not.” (Johannesson, 1995, p489) In health care, the CBA-CEA equivalence debate has been developing since the original discussion of the issue by Phelps and Mushlin (1991). A major part of the debate has been about the conditions under which CBA and CEA are equivalent4, the basic argument presented by Phelps and Mushlin (1991) being that CBA typically determines in advance the marginal value of a QALY whereas CEA calculates the ‘price’ (or ‘cost’) of obtaining a QALY but leaves the decision unstated. The question is then whether cost is greater or less than the marginal value, essentially rendering CBA equivalent to CEA, but requiring a source for the societal WTP for a QALY. Johannesson (1995) went further, examining more closely the conditions under which CBA is equivalent to CEA. These conditions are where the WTP per unit of effectiveness is constant and the same for all individuals. If this assumption is relaxed (e.g. when WTP per unit of effectiveness is allowed to vary with the size of the health effects) the equivalence will no longer hold. Thus, as highlighted in the above quote, WTP per QALY data may still be required, but it may also have to be more contextspecific. Garber and Phelps (1997), in discussing the welfare economic foundations of CEA, again show that CEA is equivalent to CBA under certain conditions, a crucial factor 4
Another line of debate addresses the issue from the point of view of defining economic evaluation types according to the original efficiency question to be addressed, whether one of technical or allocative efficiency (Donaldson, 1999). This is less relevant in the context of the research proposed here.
4
being the degree to which QALYs represent individuals’ preferences. Bleichrodt and Quiggin (1999) and Dolan and Edlin (2002) extend the argument further, their principal aim appearing to be to examine the conditions under which the CBA of proposed health care expenditure decisions (based on WTP) and CEA (based on QALYS) would produce consistent conclusions. While Bleichrodt and Quiggin allow the possibility that WTP per QALY gained may increase with income or wealth and derive conditions under which CEA (based on constrained QALY maximisation) would be consistent with CBA, Dolan and Edlin argue that these conditions are excessively restrictive and unrealistic and instead examine the possibility of producing a WTP per QALY gained that is constant and, in particular, independent of income/wealth. Dolan and Edlin then proceed to demonstrate the impossibility of the latter under circumstances in which the marginal utility of consumption is taken to be an increasing function of health status – a result which is, on reflection, hardly surprising. Does this render the quest for a defensible WTP-based monetary value of a QALY doomed to failure? Arguably, the answer is firmly in the negative. In particular, seeking an individual WTP-based monetary value of a QALY that is independent of income or wealth is about on a par with the aspiration that individual WTP for a reduction in the risk of premature death would be independent of income or wealth when both theory and empirical evidence point strongly in the direction of the conclusion that safety is a strictly normal good. But this does not prevent the UK Department for Transport, as well as other Government departments in this country and abroad form employing WTP-based values for the prevention of a statistical premature fatality (VPF) that are based on central tendency measures (typically arithmetic means) of the population distribution of individual WTP for risk reduction. The fact that these VPFs are then applied uniformly to all groups in society whatever the income levels of members of the group clearly entails the implicit use of inverseincome distributional weights. Furthermore, it should not be forgotten that it has been shown recently, at least theoretically, that QALYs also ‘suffer’ from not being independent of income/wealth (Donaldson et al., 2002) – also, see footnote 3. Similar arguments apply to the relationship between WTP-based values of a QALY and age. It would therefore appear that not only are all economics-based valuation methods subject to the ‘distributional problem’ but also that there is already a precedent in UK public-sector decision making for applying distributional weights to individual WTP for reductions in risks to life in order to arrive at an overall value in the form of a population mean (or possibly median) of individual values. This represents the state of the art of dealing with any such baises. 2.1.2
Empirical estimates of WTP for a QALY
One hundred and forty-four empirical applications of WTP in a health care context covering the period 1981-2001 inclusive were identified and reviewed. Of these, only four reported data on WTP for a QALY (Blumenschein and Johannesson, 1998; Cunningham and Hunt, 2000; Olsen and Donaldson, 1998; Zethraeus, 1998). A fifth study provided sufficient data from which to calculate WTP for a QALY estimates, but did not present these estimates in the paper (Bala et al., 1998).
5
Three of the five studies elicited values from patients (Cunningham and Hunt, 2000; Blumenschein and Johannesson, 1998; Zethraeus, 1998) while the other two elicited values from the general public (Bala et al., 1998; Olsen and Donaldson, 1998). In the Cunningham and Hunt (2000) study, 40 patients about to start orthognathic treatment were asked to value their own health status by means of a standard gamble (SG) question. They were also asked their WTP for a treatment to correct their dentofacial deformity. A payment scale was used to elicit WTP values. WTP for a QALY was estimated using the mean health state utility value (0.73) and the mean WTP (£6833). Assuming a post-treatment health state utility value of 1.0 and a remaining life expectancy of 50 years, the mean QALY gain from treatment was estimated to be (1.0 – 0.73) x 50 = 13.5. This was divided into the mean WTP of £6833 to give a WTP for a QALY of £506. Blumenschein and Johannesson (1998) asked 69 patients with asthma to value their own health status using a rating scale (RS), time trade-off (TTO) and SG. In addition, patients were asked their WTP for a cure for asthma. WTP values were elicited using both dichotomous choice and bidding game approaches. Mean health state utilities for the RS, TTO and SG were 0.68, 0.89 and 0.91 respectively. Mean WTP was US$189 per month for the bidding game and US$343 for the dichotomous choice method. The authors state that the implied WTP for a QALY from their study lies in the approximate range US$7,000 (£4,300) to US$46,000 (£24,700)5. While no details are presented on how these estimates were arrived at, it would seem that the authors simply divided the mean annual WTP amounts by the mean expected health improvement (assuming a post treatment health state utility of 1.0). Thus, the lower and upper value in the range were calculated as (US$189 x 12) ÷ (1.0 – 0.68) = US$7,088 and (US$343 x 12) ÷ (1.0 – 0.91) = US$45,733 respectively. Zethraeus (1998) asked 104 women undergoing hormone replacement therapy (HRT) to indicate on a RS their health status before they began HRT and their current health status. Values for the same two health states were also elicited using TTO questions with a 30 year time period. In addition, women were asked to state their WTP for HRT using a dichotomous choice approach. The mean change in health state utilities was estimated for the RS and TTO and divided into the mean WTP for HRT to give WTP for a QALY estimates of 118,400 Swedish Krone (SEK) (£8,600) and 156,100 SEK (£11,400) for the RS and TTO respectively. In the Bala et al. (1998) study, 114 members of the general population aged 65-70 years were asked their WTP for three treatments which reduce the pain from shingles. A double bounded dichotomous choice method was used to elicit values. Health state utility values were also elicited for both severe and mild pain using the SG with a 20 year time horizon. Mean QALYs gained and median WTP for each of the three treatments were estimated, but the authors stopped short of presenting the data in terms of WTP for a QALY. Dividing the median WTP values by the corresponding mean QALY gains gives WTP for a QALY estimates of US$34,455 (£21,300) for treatment one, US$49,133 (£30,300) for treatment two and US$15,588 (£9,600) for treatment three.
5
All currency conversions are approximate and were performed using exchange rates at 12/2/03
6
In the study by Olsen and Donaldson (1998), 150 members of the general public were asked their WTP for three disparate health care programmes: a helicopter ambulance service; an expansion in the provision of heart operations; and an expansion in the provision of hip operations. From the descriptions of the programmes the authors estimated the expected QALY gains for each programme under three separate sets of conditions: undiscounted QALYs; QALYs discounted at 10%; and undiscounted QALYs under a conservative assumption that the quality enhancement part of the QALY gain is halved. Mean WTP for the three programmes was 316 Norwegian Krone (NOK) for the helicopter ambulance, 306 NOK for the heart programmes and 232 NOK for the hip programme. Dividing these values by the expected QALY gains gave a WTP for a QALY range of 0.2 NOK (£0.02) to 6.7 NOK (£0.60). The estimates of WTP for a QALY in the above studies are very diverse, ranging from £0.02 to £30,300. This alone casts significant doubt on the plausibility of using data from previously published studies. This doubt is compounded by the fact that none of the studies was designed with the specific purpose of directly estimating WTP-based values for a QALY. In addition, only one study reports values from the UK, and these are patients’ values, not those of society (Cunningham and Hunt, 2000). Of the two studies which elicited societal values, one of them presents extremely implausible WTP estimates of less than £1 for a QALY (Olsen and Donaldson, 1998). There is a sixth study, by Hirth et al. (2000), which contains WTP-for-a-QALY estimates based on methods from the value-of-life literature – therefore, this method is dealt with below (see section three). As well as the above studies, the literature review identified a further 13 studies which reported health status values and WTP data alongside one another, but did not attempt to estimate WTP for a QALY (Barrett et al., 1994; Coley et al., 1996; Kobelt, 1997; Krabbe et al., 1997; Kupperman et al., 2000; Lundberg et al., 1999; O’Brien and Viramontes, 1994; Smith, 2001; Sorum, 1999; Swan et al., 1997; Thompson, 1986; Voruganti et al., 2000; Zethraeus et al., 1997). Twelve of these did not convert the health status values into QALYs, while the other study presented quality adjusted life days (Swan et al., 1997). Although to date, there is no study in the discrete choice experiment (DCE) literature which has produced a WTP-based value of a QALY, this is an area of research which has experienced substantial growth in the health economics field in recent years and for which there is much potential for estimating such a value. DCEs6 are an attributebased stated preference valuation technique. They draw upon Lancaster’s economic theory of value (Lancaster, 1966; 1971) and random utility theory (McFadden, 1973; Hanemann, 1984). Attributes of the commodity being valued are first defined, and levels assigned to them. Statistical design theory is then used to construct efficient choice sets and individuals are presented with these choices in a context that is familiar to them (Zwerina et al., 1996). Analysis of the choice data allows estimation of the relative importance of the separate attributes, the marginal rates of substitution (MRS) between attributes and, if a price proxy is included, WTP for both changes in
6
Choice experiments are a subset of a broader experimental technique known as conjoint analysis. Early applications of DCEs in health referred to the technique as conjoint analysis.
7
individual attributes, as well as changes in any combination of attributes (Louviere et al., 2000). In a recent review about half the DCEs had been applied within a health economic evaluation framework, many to value dimensions of health benefit beyond health outcomes (Ryan and Gerard, 2003). Other areas of application include: health insurance planning and premium setting, understanding labour supply characteristics and agency relationships, extracting time preference values, developing prioritisation frameworks. Whilst it was noted in the late 1990s that a potentially exciting area of research would be the application of the DCE approach to estimate WTP for a QALY (Ryan, 1999), to date very little research has investigated this. Hakim and Pathak (1999) compared the rating scale, standard gamble and discrete choice approach for measuring EuroQol health state preferences. They compared the convergent validity and predictive validity, and concluded that future research was necessary to compare preference scores of the different methodologies. To date this has not been done. Two studies have used the DCE approach to estimate preferences for specific health outcome measures: Netten et al. (2000) used the DCE approach to develop a preference-weighted utility scale for measuring the outcome of social care and Johnson et al. (2000) employed the DCE approach to estimate WTP for improved respiratory and cardiovascular health using a modification of the Quality of Well Being (QWB) index. 7 Bryan et al. (2002) used the DCE approach to assess the importance of “number of people” and “chance of success” in addition to “quantity” and “quality of life”. In summary, the DCE approach has not been applied to the specific research question addressed in this protocol. However, its potential for doing so has been highlighted by its use in the above-mentioned studies. 2.1.3 Beyond QALYs: is there value in other characteristics of health care or its beneficiaries? Previous research has shown that respondents have preferences over non-health characteristics of health care and the severity of illness of beneficiaries of health care (see below), with weaker evidence having shown that it matters whether QALYs are made up of quality-of-life gains alone or gains in survival combined with quality of life. Therefore, it is important to ‘contextualise’ valuations of health gains by taking such factors into account, or, at least, to know how important they are. The first and third of these constitute benefits which emanate directly from the care provided. The second issue is more related to broader “societal” concerns, which is more about who receives care as opposed to the magnitude of the benefit. Other societal aspects, such as the age of beneficiaries and notions of “culpability” (as mentioned in the initial call by NCCRM and NICE), can also be considered, building on a particular advantage of the DCE approach used in the proposed research. Of course, in addition, we will test for the impacts of respondent characteristics on valuations obtained. It may also be thought that valuation methods in health care should take account of externalities in the sense that people care for others and, therefore, derive benefit in 7
The literature review identified a study by Harwood et al. (1994) which had used conjoint analysis more broadly, employing a rating exercise to develop a handicap measurement scale.
8
seeing them use health care when needed. However, although the “caring externality” is argued to be one of the sources of market failure in health care (Culyer, 1971), and thus part of the justification for NHS-type systems (Evans, 1984), this does not necessarily mean that such benefits should be accounted for in economic evaluation of such publicly-funded goods – see Appendix A below. Therefore, this aspect is not addressed in the empirical part of the proposal. However, this does not preclude investigation of a societal perspective, with the respondent taking on the role of social decision maker, or “setting him/herself aside”, as opposed to the self-only concept of externalities. Does ‘process’ matter? More generally, the question here is whether scenarios describing health care programmes to be valued should contain information which is additional to the health gains involved and, if they do contain such information, does it have an impact on values obtained. Mooney (1994) and Gerard and Mooney (1993) have made a comprehensive case for inclusion characteristics of health care, beyond health itself, in the valuation process. Generally, studies in the health economics literature demonstrate that informational aspects beyond health gains are important (Berwick and Weinstein, 1985; Ryan and Hughes, 1997; Ryan and Farrar, 2000), although the results have not always been as expected, in terms of whether additional information will have a positive or negative effect on valuations (Donaldson and Shackley, 1997) or the magnitude of effect of additional information (Protière et al., 2003). The last of these is, experimentally, the most rigorous study, and, despite the implausibly large proportion of additional WTP attributed to non-health information, the study obviously shows that such information has an effect. The ‘rule of rescue’: It will be seen below that studies have demonstrated that respondents to WTP surveys face challenges in understanding, and therefore valuing, small risk reductions. In estimating a monetary value of a QALY, therefore, it is tempting to arrive at such a value through having respondents value gains in quality of life only. However, this raises a different issue in that valuations of equivalent QALY gains are likely to be different if made up of life-saving/extending improvements as well as quality-of-life gains as opposed to being composed of quality-of-life gains alone. Hadorn (1991) has made a similar point in his discussion of possible reasons for difficulties in implementing the cost-per-QALY results from the Oregon experiment in health care priority setting. It has also been shown that WTP for a QALY is lower for interventions which enhance quality of life only, although that study was not set up to examine that particular issue and other factors may have led to this result (Olsen and Donaldson, 1998). Severity of illness and other ‘societal’ concerns: Although the issue of severity has not been the subject of much research, a landmark study has shown how it would be possible to adjust QALYs for, amongst other things, the severity of a potential beneficiary’s pre-treatment condition (Nord et al., 1999). The suggested method for achieving this was through person trade-offs, although it may also be possible to similarly adjust monetary measures of value, as Nord et al. point out. Furthermore, cases have been made for adjusting QALYs according to various factors, such as age (Williams, 1997; Tsuchiya, 1999; 2000; Dolan, 2000; Dolan and Cookson, 2000), and some other more controversial factors mentioned in the original call for proposals from NCCRM/NICE.
9
2.2
Lessons from the safety literature
Although the concept of WTP has existed for a long time, it was not until the 1960s that the first empirical application was published in a journal (Davis, 1963). This was in the area of environmental policy evaluation, and was specifically concerned with estimating the benefits of outdoor recreation in a backwoods area in the US state of Maine. During the 1970s, the method was further developed in studies of the valuation human life, as applied to safety and transport policies (Jones-Lee, 1974, 1976; Mooney, 1977). The latter area of work is of greater direct relevance to the issue addressed in the research proposed in this protocol. This work has involved the development of methods to elicit a “value of statistical life” and, thus, is more closely related to the challenge of valuing a QALY (or “healthy year”). Given its subject matter, it is also one of the most controversial issues in the appraisal of proposed public sector projects. Because of the importance of developments in this literature to one of the main research methods outlined in this proposal, a fuller review is provided in Appendix A, with the main issues (which duplicate some of Appendix A) summarised here. Until the 1980s most countries that explicitly addressed the public sector safetyvaluation issue tended to use some variant of the so-called “gross output” or “human capital” approach. Under this approach the primary component of the “cost” of the premature death of an individual is treated as the discounted present value of that individual’s future output extinguished as a result of his or her premature demise. In some countries (including the UK) a further more-or-less arbitrary allowance was then added to the gross output figure to reflect the “pain, grief and suffering” of the victim and/or his/her surviving dependents and relatives. Values for the prevention of premature death are then defined in terms of the costs avoided. To give an example of the costs and values that emerge under the gross output approach, the UK Department for Transport’s most recent gross output – based value for the prevention of a road fatality – based on national averages – was £180,330 in 1985 prices, of which about 28% was an allowance for pain grief and suffering. Updated for inflation and growth of real output per capita this figure would now stand at some £460,000 in 2002 prices. Not surprisingly, many economists have objected to the gross output approach on the grounds that most people almost certainly value safety largely because of their aversion to the prospect of their own and others’ death and injury as such, rather than because of a concern to preserve current and future levels of output and income (Schelling, 1968; Mishan, 1971; Jones-Lee, 1989). Given this, it has been argued that values of safety ought ideally to be defined so as to reflect people’s “pure” preferences for safety, per se, rather than in terms of effects on output and income, as in the gross output approach. However, in order to define and estimate values of safety in this way we clearly require some means of measuring people’s preferences for safety and, more particularly, their strength of preference. How can one do this? Arguably, the most natural measure of the extent of a person’s preference for anything is the maximum amount that he or she would be willing to pay for it. This amount reflects not only the person’s valuation of the desired good or service relative to other
10
potential objects of expenditure, but also the individual’s ability to pay – which is itself a manifestation of society’s overall resource constraint. So, under what has naturally come to be known as the “willingness-to-pay” (WTP) approach to the valuation of safety, one first seeks to establish the maximum amounts that those affected would individually be willing to pay for (typically small) improvements in their own and others’ safety. These amounts are then simply aggregated across all individuals to arrive at an overall value for the safety improvement concerned. The resultant figure is thus a clear reflection of what the safety improvement is “worth” to the affected group, relative to the alternative ways in which each individual might have spent his or her limited income. Furthermore, defining values of safety in this way effectively “mimics” the operation of market forces – in circumstances in which markets typically do not exist – insofar as such forces can be seen as vehicles for allowing individual preferences to interact with relative scarcities and production possibilities to determine the allocation of a society’s scarce resources. In order to standardise values of safety that are derived from the WTP approach and render them comparable with values obtained under other approaches (such as gross output), the concept of the prevention of a “statistical” fatality or injury is applied. To illustrate this concept, suppose that a group of 100,000 people enjoy a safety improvement that reduces the probability of premature death during a forthcoming period by, on average, 1 in 100,000 for each and every member of the group. The expected number of fatalities within the group during the forthcoming period will thus be reduced by precisely one and the safety improvement is therefore described as involving the prevention of one “statistical” fatality. Now suppose that individuals within this group are, on average, each willing to pay $w for the 1 in 100,000 reduction in the probability of death afforded by the safety improvement. Aggregate WTP will then be given by $w x 100,000. This figure is naturally referred to as the WTP-based value of preventing one statistical fatality (VPF) or alternatively as the value of statistical life (VOSL). Clearly, in the above example, average individual WTP, $w, for the average individual risk reduction of 1 in 100,000 is a reflection of the rate at which people in the group are willing to trade off wealth against risk “at the margin”, in the sense that the trade-offs typically involve small variations in wealth and small variations in risk. Empirical work on the valuation of safety thus tends to focus upon these individual marginal wealth/risk trade-off rates. On a somewhat more cautionary note, it is extremely important to appreciate that, defined in this way, the VPF is not a “value (or price) of life” in the sense of a sum that any given individual would accept in compensation for the certainty of his or her own death – for most of us, no finite sum would suffice for this purpose, so that in this sense life is literally priceless. Rather, the VPF is aggregate WTP for typically very small reductions in individual risk of death (which, realistically, is what most safety improvements really offer at the individual level). Similarly, in the case of trying to estimate a societal WTP-based monetary value of a QALY, as argued in sub-section 3.3, this is probably also most appropriately viewed as a group-aggregate WTP for marginal gains in quality of life or life expectancy given that, at least in the case of a randomly-selected sample of the public, such gains will typically be marginal, though
11
of course the same cannot necessarily be said for those already suffering from health impairments. A similar argument, about taking an insurance-based approach to valuing publicly-provided health care whereby individual members of the community are informed about the probability of needing care as well as the probability of it being successful, has been made in the health economics literature by Gafni (1991) and O’Brien and Gafni (1996). But how, in fact, are WTP values of safety estimated in practice? Broadly speaking, three variants of empirical estimation procedure have been employed to derive WTPbased values of safety. These are known respectively as the “revealed preference” (or “implied value”), the “contingent valuation” (or “expressed value”) and “relative valuation” approaches. Basically, the revealed preference approach involves the identification of situations in which people actually do trade off income or wealth against physical risk – for example, in labour markets where riskier jobs can be expected to command clearly identifiable wage premia (Smith, 1983; Viscusi and Moore, 1989). By contrast, the contingent valuation approach involves asking a representative sample of people more or less directly about their individual WTP for improved safety, (or, sometimes, their willingness to accept compensation for increased risk). The difficulty with the revealed preference approach when applied to labour market data is that it depends on being able to disentangle risk-related wage differentials from the many other factors that enter into the determination of wage rates. The approach also presupposes that workers are well-informed about the risks that they actually face in the workplace. In addition, those whose jobs do carry clearly identifiable wage premia for risk may not be representative of the work force as a whole, in that such people almost certainly have a below-average degree of risk-aversion (Gegax et al., 1991). The great advantage of the contingent valuation approach is that it allows the researcher to go directly and unambiguously to the relevant wealth/risk trade-off – at least, in principle. On the other hand, the contingent valuation approach has the disadvantage of relying upon the assumption that people are able to give considered, accurate and unbiased answers to hypothetical questions about typically small changes in already very small risks.8
8
Equally contentious to placing monetary values on safety and health policies has been that area of environmental economics research involving assessment of so-called “non-use” and “existence” values. It is in this area that most work has been conducted on the extent to which values derived in such contingent-valuation surveys reflect “real-world” behaviour. In the early 1990s, the literature on this issue was split, with five studies showing WTP values elicited from surveys to be greater than those from real behaviour whilst five studies gave consistent results (Hanemann, 1993). Since then, Carson et al. (1996) have shown in a systematic review of the literature that, compared with revealed preference methods, contingent valuation WTP estimates tend to be lower (on average 0.89 of revealed preference estimates), which is the opposite of what many people might expect. One study in the health economics literature has shown that a revealed preference and contingent valuation method arrive at similar valuations (Kennedy, 2002), whilst others have shown the opposite to the Carson et al. review (Clarke, 1997), with one also making the claim that it may still be possible to correct for any overestimation (Blumenschein et al., 2001).
12
By contrast, unlike the revealed preference and contingent valuation approaches, the relative valuation approach does not involve an attempt to estimate wealth/risk tradeoff rates directly, but rather seeks to determine the value of preventing one kind of physical harm relative to another. Thus, for example, the UK Department for Transport’s current monetary values for the prevention of non-fatal road injuries of various levels of severity were obtained by applying estimates of such relative valuations to an absolute monetary “peg” in the form of the Department’s existing WTP-based roads VPF. All of this having been said, it is important to note that both quantitative and qualitative research has cast doubt on the reliability and validity of WTP values for safety derived through the above direct contingent valuation method. As well as sequencing and framing effects, a prominent issue has been the lack of ability of the method to account for embedding and scope. That is, respondents tend to view safety improvements as a “good thing” and, therefore, will often state much the same WTP for different sizes of risk reduction, whether for fatal or non-fatal injuries (Jones-Lee et al., 1995; Dubourg et al., 1997; Beattie et al., 1998). It may be unreasonable to expect respondents to give accurate answers to hypothetical questions which involve direct trade-offs between wealth and small reductions in risk. Therefore, Carthy et al. (1999) have suggested a less-direct CV/SG chained approach which breaks down the valuation process into a series of manageable steps which involves chaining together responses to WTP and SG questions. First, respondents are presented with a question asking them about their WTP for the certainty of a complete cure for a given non-fatal road injury and their willingness to accept compensation for the certainty of remaining in the impaired health state (the combination of which, based on some reasonable assumptions about underlying preferences obeying minimal conditions of consistency and regularity, it is argued, gives a reasonable estimate of the marginal rate of substitution of wealth for the risk of the non-fatal injury). Second, respondents are presented with a SG question aimed at determining the ratio of the health state value for death over that for the non-fatal injury. The monetary value from the first stage can then be combined with the ratio from the second stage to obtain a WTP for reduced risk of death. In fact the current Department for Transport VPF was derived using this approach and currently stands at £1.19 million in June 2001 prices9, with corresponding values for the prevention of serious and slight non-fatal road injuries estimated using the relative valuation approach being £134,190 and £10,350 respectively. The CV/SG chained approach is, perhaps, more realistic in that most people can relate to giving a monetary value for avoiding a non-fatal injury of the sort they are likely to have experienced, and people are not asked directly to place a monetary value on a small risk reduction. The method has shown promise in terms of being subject to less marked embedding effects and other biases than earlier approaches (Beattie et al., 1998). It should be also be noted here that another major focus in the environmental literature with respect to validity has been on scope effects: especially whether split samples of respondents willing to pay more for greater amounts of the good being valued, as one would expect. Carson (1997) shows that most studies (31 of 35 reviewed) reveal sensitivity to scope. It should also be noted that it has been argued 9
Adjusted to 2002 prices this figure would be around £1.25 million, as quoted on page 1.
13
that DCEs can overcome the scope problem as they force respondents to think more about the individual attributes of the commodity being valued (Hanley et al., 2002). Despite such positive results, doubts still remain about WTP values elicited through hypothetical surveys, and scope tests have taken on the status of being the “acid test” for any particular study. Also, this has led to more qualitative methods being used to examine the thought processes underlying respondents' stated values (Schkade and Payne, 1994; Chilton and Hutchison, 1999), a trend which will likely (and justifiably) continue. Given this potential, the CV/SG chained approach will form the basis of one of the main methods for deriving a WTP-based value of a QALY. The theoretical case for this, along with a case for one other contingent-valuation based method, is made in section three. In making this case, approaches based on adapting existing values to estimate a WTP-based value of a QALY are also outlined, and their limitations discussed. These values do, however, provide a useful basis for comparison with values generated from the two new contingent-valuation-based methods proposed in section three and the DCE approach introduced in the following sub-section and in section four. 2.3
Summary of issues to be addressed
2.3.1. The basic approaches All of the above points to the need to estimate more directly a WTP-based value for a QALY, something which has not been derived explicitly in previous studies as reported in the health economics literature. The method should use up-to-date methods aimed at overcoming some of the cognitive problems associated with deriving estimates of WTP. In view of the promise shown by the CV/SG chained approach, it would seem to form a natural candidate as a basis for estimating a WTPbased monetary value of a QALY and our suggestions concerning the precise manner in which it might be applied are detailed in the following section. This method does appear to have the potential to overcome some important problems: first, it breaks down the valuation process into a series of (hopefully) more manageable tasks; second, despite commencing with questions asking respondents to value certain gains in quality of life, the last stage of the chaining approach ensures that links to uncertainty and life years are still incorporated. However, we are conscious that there are also a number of other possible approaches to deriving a WTP-based value of a QALY based on the contingent valuation method. Although these are also outlined in the following section, cases are made for the CV/SG chained and WTP for gains in life expectancy approaches.10 Although potential for deriving estimates of WTP for a QALY by use of DCEs has been demonstrated, this has not been done directly. The principles of how this might be done are outlined in section four, where the potential for using DCEs to take into account broader societal factors (listed again in 2.4.2) is also displayed. 10
Ultimately, however, the pros and cons of the various possible approaches – and possibly others that have not occurred to us – are matters for whoever carries out the research to explore and evaluate. Indications of how this might be done are therefore in our view, a vitally important aspect of any response to our suggested protocol.
14
2.3.2
Additional important issues
It is also important to account for important theoretically-relevant, informational and societal aspects which have been shown in health and other applied literatures either to have an impact on values or, as in the case of scope effects/embedding, to be both controversial and important to test for. In addition to scope tests, these aspects include the information presented to respondents, age, health status, accounting for both quantity as well as quality of life, and adjusting values for broader societal factors such as age, initial health state (or severity), culpability etc.. Experiments designed for addressing these issues are outlined in section five. It is important to address aspects of the validity of the approaches by the use of carefully constructed qualitative studies prior to, and alongside, surveys and to assess plausibility by comparing results obtained with those obtained through the use of similar methods in non-health fields, especially (in the case of the methods proposed in this protocol) valuations of life derived from studies of safety improvements. There are no separate sections on either of these; rather, how these considerations fit into the proposed research is outlined when describing the survey methods in section six.
15
3.
ESTIMATING
A WTP-BASED MONETARY VALUE OF A QALY BY THE CONTINGENT VALUATION METHOD: AN ASSESSMENT OF SIX POSSIBLE APPROACHES AND A SUGGESTED WAY FORWARD
3.1
Simplifying assumptions
For the sake of simplicity, in what follows in this section the discount rate will be treated as being zero. It would, however, be a straightforward matter to introduce appropriate discount factors to reflect non-zero discount rates. In addition, it will also be assumed that the appropriate WTP-based value for the prevention of a (statistical) premature fatality (VPF) is the current Department for Transport roads figure updated to 2002 prices, which would be about £1.25 million. This would correspond to a mean individual WTP of £25 for a 2 in 100,000 reduction in the risk of death in a car accident during the coming year. Finally, again in the interest of simplicity, the argument will take as its reference point an individual aged 35 with 40 years of remaining life expectancy and it will be assumed that all of these 40 years will be spent in full health i.e. that there will be no deterioration in the quality of life as the individual ages. Clearly it would, in principle, be possible to modify the argument so that it dealt with an age distribution that was representative of the population as a whole and to allow for the possibility of a deterioration in the quality of life in later years. Prima facie, there would then appear to be six possible approaches to the estimation of a WTP-based monetary value of a QALY 11 based on the contingent valuation method. The first four of these approaches are essentially “back-of-the-envelope” calculations based on WTP-based values of safety that already exist and are offered as a means of generating “marker” estimates with which values derived under the fifth and sixth approaches (which we view as particularly promising ways in which to proceed and would involve generation of new data via the appropriate contingent valuation surveys) could be compared. The reader who wishes simply to gain a general feel for the proposed ways forward, which essentially involves the fifth and sixth approaches in this section and the DCE approach introduced in the next section, may wish to skip directly to our discussion of these approaches (commencing on page 20). 3.2
Six possible approaches
Approach 1 The VPF of £1.25 million is essentially aggregate WTP across a large group of people for a risk reduction that will entail the expected prevention of one premature death during the coming year. For a representative 35 year old this prevention of premature death would entail the avoidance of the loss of 40 years of life expectancy.
11
Note that, for the purposes of expediency, ‘QALY’ and ‘life year’ are used interchangeably within this section of the paper.
16
Assuming that each year is equally valued and a zero discount rate, this entails that the WTP-based monetary value, V1, of each year is given by: V1 =
£1.25 x 10 6 40
= £31,250
(1) (2)
Under this approach the WTP-based monetary value of a QALY would therefore be treated as £31,250. By contrast, if the discount rate was taken to be 5% then the annualised sum that would have a discounted present value of £1.25 million over 40 years would be £73,450.12 In fact, this is essentially the approach that has been adopted in those attempts that have been undertaken to date to estimate the WTP-based monetary value of a QALY – see for example: Hirth et al. (2000) and Hurley et al. (2000) - though it should be noted that in the case of the former the discount rate was taken to be 1.5% and in the latter 5%. In addition, in the latter, quality of life adjustments were also applied to different ages. Approach 2 Denoting life expectancy in years by Ey and the probability of death during the coming year by p, it is a straightforward matter to show that:
dE y dp
≈ Ey.
(3)
(For a demonstration of this result, see Jones-Lee (1976) p.142). It follows that the gain in years of life expectancy, ∆Ey, resulting from a 2 x 10-5 reduction in the risk of death during the coming year for our representative 35 year old with 40 years of remaining life expectancy is well-approximated by: ∆Ey
= 40 x 2 x 10-5
(4)
= 8 x 10-4
(5)
which is equivalent to a gain of 0.292 days. Given that a VPF of £1.25 million entails mean individual WTP of £25 for a 2 x 10-5 reduction in risk of death during the coming year, there are then two possible ways in which to infer a WTP-based value of a QALY.
12
In this estimate, life years are not discounted.
17
Approach 2a First, on the basis of the somewhat implausible assumption of linear proportionality in the relationship between individual WTP and increase in life expectancy, one 365 would simply multiply £25 by to obtain an individual WTP-based value of a 0.292 QALY, V2 as:
V2 = £31,250
(6)
which is, of course, the same monetary value of a QALY as given by Approach 1. Approach 2b
Alternatively, one could aggregate WTP for a 0.2.92 days gain in life expectancy 365 across different individuals to obtain precisely the same monetary value of a 0.292 QALY as given by Approach 2a. It is, however, vitally important to appreciate that the logic underpinning this alternative group aggregate WTP approach is fundamentally different from that employed in the linear proportionately-adjusted “individualistic valuation” Approach 2a. Approach 3
Assuming that the £25 mean individual WTP for a 2 x 10-5 reduction in the risk of immediate premature death for our representative 35 year old is effectively WTP for a 2 x 10-5 reduction in the risk of losing 40 years of remaining life expectancy, on the assumption that each year is equally valued and a zero discount rate, it follows that individual WTP, W, for a 2 x 10-5 reduction in the risk of losing each one of the remaining 40 years is given by: W=
£25 . 40
(7)
This means that aggregate WTP, V3 across 50,000 people for a 2 x 10-5 reduction in the risk of losing one year of life expectancy is given by: V3 =
£25 x 50,000 40
= £31,250.
(8) (9)
However, a 2 x 10-5 reduction in the risk of losing one year of life expectancy for each person in a group of 50,000 entails an expected saving of 50,000 x 2 x 10-5 = 1 year of life expectancy. Thus under this approach the WTP-based value of a QALY would, as under Approach 1 be £31,250. It is, however, again very important to appreciate that the value obtained under Approach 3 is – as under the second variant of Approach 2 – an essentially group aggregate WTP figure.
18
Approach 4
It is important to appreciate that both Approaches 1 and 3 rely upon a rather questionable assumption, namely that willingness to pay for a reduction in the risk of immediate premature death is no more and no less than willingness to pay for the preservation of a given number of equally-valued future life years (e.g. for our 35 year old, 40 more years). However, it may well be the case that many people’s WTP to reduce mortality risk depends on a great deal more than future life-span and could, for example, be substantially influenced by considerations such as the emotional costs of premature death to those who would be bereaved; the will to live and a concern about failing to achieve specific lifetime aspirations, such as the desire to see one’s children and grandchildren grow up. In view of this, Loomes (2002) suggests that, at least from middle age onwards, it might be more appropriate to model the relationship between individual WTP for safety as compromising a lump sum (reflecting the value of living per se) and a component that depends on remaining life expectancy. Clearly, the simplest version of such a model would take the form: Mi = α + βEi + ui
(10)
where Mi is the ith individual’s marginal rate of substitution of wealth for risk of death during the coming year in £ sterling (the population mean of which is taken to be the VPF under the WTP approach – see, for example, Jones-Lee (1989), Ei is the ith individual’s remaining years of life expectancy and ui is a random error term. Considering the Mi-versus-age relationship estimated in three UK studies (namely Jones-Lee et al. (1985), Jones-Lee et al. (1995) and Carthy et al. (1999)) which is essentially of an inverted – U shape, peaking in middle age, as a first approximation it M would seem appropriate to take the average of i - where M is the population mean M of Mi – as being about 0.5 for a 75 – 80 year old and about 1.1 for a 35 year old. Given this, taking M = £1.25 million (the current DfT VPF updated to 2002 prices), assuming that remaining life expectancy for a 75 – 80 year old is effectively zero and a zero discount rate with no adjustment for declining quality of life in later years, would give, from equation (10) Ei = 0 => Mi = α = 0.5 x £1.25 million = £625,000 Ei = 40 => Mi = α + β x 40 = 1.1 x £1.25 million
(11) (12) (13)
So that from (12) and (13):
β =
0 .6 x £1.25 million 40
(14)
19
= £18,750
(15)
Hence, substituting from (12) and (15) into (10) gives: Mi = 625,000 + 18,750 Ei,
(16)
from which it follows immediately that: dM i = £18,750 dE i
(17)
so that aggregate WTP, V4, across a large group of people for marginal gains in life expectancy that total to one year is given by: V4 = £18,750.
(18)
Under Approach 4, then, the WTP-based value of a QALY would be only a little over a half of the figure implied by Approaches 1–3. Approach 5
Under this approach one would not rely upon a predetermined monetary VPF, but would instead ask a representative sample of the population how much each person (or alternatively each household) would be willing to pay per annum over remaining lifetime for a one month (or three months, or six month) gain in life expectancy in normal health. In our experience, this kind of approach is relatively straightforward, easy to administer and produces broadly credible results. Thus, for example, in a recent pilot study carried out as part of a DEFRA-funded project on the value of reductions in air pollution carried out by a team from Durham, Newcastle, UEA and Queen Mary, University of London, respondents were asked about annual (and hence lifetime) household WTP for (a) a one month, (b) a three months and (c) a six months gain in life expectancy in normal health resulting from 10%, 25% and 50% reductions respectively, in ongoing levels of PM10 (particulate) air pollution, though it should be noted that each respondent was presented with only one of these gains in life expectancy, specific gains being selected on a random basis. With extreme outliers removed, sample mean household annual WTP responses were (approximately) £35 for the one month gain, £60 for the three month gain and £80 for the six month gain. With a zero discount rate these figures would convert to 40 year lifetime payments of £1400, £2400 and £3200 respectively. Clearly, to the extent that the marginal reduction in the coming year’s risk of premature death that yields a VPF of £1.25 million entails less than a day’s gain in life expectancy (see discussion of Approach 2) – and indeed, that even if the 2 x 10-5 annual risk reduction was effective each year over remaining lifetime then the gain in life expectancy would be only some 5 days - then even a one month gain, let alone a three months or six months gain, can hardly be regarded as marginal, so that the WTP
20
responses in the DEFRA project were almost certainly heavily influenced by binding mental-accounting sub-budget constraints. In view of this, it would seem that there is a case for attempting to infer the WTP responses that people might be expected to give if (a) the gain in life expectancy was genuinely marginal and (b) they fully understood how such a gain in life expectancy would be effected i.e. by small reductions in the annual risk of death. (Note that in the absence of such an understanding, most respondents would almost certainly treat the marginal gain in life expectancy as being merely the “add on” of a few hours or days at the end of life in, say, 40 years time and therefore being of little, if any, value to them.) Accordingly, it would seem appropriate to fit some form of non-linear, increasing concave function to the DEFRA project WTP responses converted to per capita, rather than household, values. Given an average UK household size of 2.5, the lifetime responses covert to per capita lifetime payment figures of £560, £960 and £1280 for one, three and six month gains in life expectancy. Assuming an underlying WTP (£W)/gain in days of life expectancy (G) relationship of the form: W = α ln(G + 1) ,
(19)
which is strictly increasing, concave and such that W = 0 when G = 0, the simplest fit of W = 213 ln(G + 1) which passes through the mean of W and the mean of ln(G + 1) - would appear to be a not unreasonable first approximation, yielding: G = 30 => W = £730.
(20)
G = 90 => W = £970
(21)
G = 180 => W = £1107
(22)
and
which seems like a fairly good fit. Setting G = 5 (i.e. a five day gain in life expectancy) then yields W = £382. This then 365 entails that aggregated across = 73 individuals, overall willingness to pay, V4, 5 for an expected total gain of one year of life expectancy would be given by: V5 = £27,860
(23)
Under this approach, then, the WTP based value of a QALY would be £27,860, which is of the same broad order of magnitude as yielded by Approaches 1, 2 and 3.
21
Approach 6
This is based on the CV/SG chained approach described in Carthy et al. (1999) and would involve presenting respondents with the following three questions: Q1
Q2.
Q3.
Please rate your current state of health. Suppose that you were diagnosed as suffering from a condition which will last for about a year and which would result in you being in the following health state,…(Choose condition that involves a loss of health-state utility of 0.01).13 Suppose also that condition X cannot be treated under the NHS, but that there is a treatment available privately that will completely cure the condition. What is the maximum amount that you would be willing to pay for the treatment? Response A Suppose you were diagnosed as suffering from the same condition, but that there is no treatment that would cure the condition, and so you cannot return to your current state of health. Suppose also that on the same day as the diagnosis you were to enjoy an unexpected gain in wealth in the form of a win on the National Lottery or a tax rebate. How large would the gain have to be to just “make up for” the diagnosed condition and leave you feeling that it had been on balance “neither a good day nor a bad day”? Response B Suppose again that you were diagnosed as suffering from the same condition, but that a treatment was available on the NHS. This treatment, if successful, would completely cure the condition, and restore you to your current state of health. However, there is a chance that, though the treatment would cure the condition, it could have adverse side effects which, if they occurred, would involve the loss of one year’s life expectancy. At what probability of treatment failure would you decide against having the treatment? Response C
Value of gain of one QALY, V6, resulting from an improvement in the quality of life (rather than a gain in life expectancy) is then inferred as: V6 = 100 x Response A.
(24)
“Cost” of loss of one QALY, C6, resulting from deterioration in the quality of life is inferred as: C6= 100 x Response B.
(25)
13
Current health will be rated using the EQ-5D set of quality-of-life descriptors and, thus, will be scored using the EQ-5D UK tariff (Dolan et al., 1995, Brooks , 1996). The interviewer will then use a programme to identify the health state in the EQ-5D system which results in a 0.01 reduction in quality of life relative to the respondent’s current health state. The reasons for using this approach are (a) it involves identifying realistic changes in health which have been shown empirically to result in such a ‘utility’ loss and (b) changes in health are characterised in terms of descriptors commonly used in empirical studies (i.e. the EQ-5D). To test for scope, larger changes should also be identified. A range of such changes, from a review of technologies which have been reviewed by the National Institute of Clinical Excellence, is documented in Appendix B.
22
(Notice that V6 and C6 are best viewed as being aggregates across a group of 100 individuals, rather than individualistic values inferred on the basis of the doubtful assumption of linear proportionality in the WTP/QALY-gain relationship). In turn, the marginal rate of substitution, MX, of wealth for risk of condition X can be inferred as a weighted average of Response A and Response B given assumptions concerning the general properties of underlying conditional utility of wealth functions – see Carthy et al. (1999). As a rough and ready rule of thumb, MX can be taken to be in the region of MX = λ Response A + (1 - λ ) Response B where λ ≈ 0.8. Given that the ratio of loss of utility from loss of one year of life expectancy ÷ loss of utility from 1 condition X is equal to - see Carthy et al. (1999), it can then be shown Response C that the marginal rate of substitution, MY, of wealth for risk of losing one year of life expectancy is given by MY =
MX . Response C
(26)
MY can then be interpreted as aggregate WTP across a large group of people for reductions in the individual risk of losing one year of life expectancy in normal health, where the expected aggregate gain in life expectancy across the group is one year. This would therefore constitute a WTP-based monetary value of a QALY in precisely the same sense as that given by Approach 3. Note that in an “ideal” world of the type implicitly assumed by the proponents of the QALY methodology, one would expect V6< MY < C6.
(27)
and, indeed, according to the argument developed in Bleichrodt and Quiggin (1999) MY = 100 MX
(28)
Whether or not this would turn out to be the case in practice remains to be seen. Alternatively, respondents could be presented with the type of “modified” standard gamble question described in Carthy et al. (1999). Essentially, this would replace the certainty of condition X under the “non-treatment” option in Q3 with the possibility of loss of one year of life expectancy. The advantage of this is that it would avoid the “certainty effect” under which “non-treatment” would be selected, however small the probability of adverse side effects under the “treatment” option. Such a question might take the following form: Q 3ˆ
Suppose that you were involved in a car accident and if left untreated would suffer from severe permanent disability. However, two treatments are available on the NHS. Treatment A would avoid severe permanent disability and would result in condition X which, as we said earlier, would last for about a year. However, while the treatment would avoid severe permanent disability and result in condition X, there is a probability θ (>o) that it could also have 23
adverse side effects which, if they occurred, would involve the loss of one year’s life expectancy. By contrast, Treatment B would completely cure you and return you to normal health with immediate effect. However, there is a probability ∏ that this treatment could also have adverse side effects, which, if they occurred, would involve the loss of one year’s life expectancy. Clearly, if ∏ were equal to θ then Treatment B is undoubtedly preferable to Treatment A. However, suppose that ∏ was larger than θ . At what level of ∏ would you decide to take Treatment A instead? Response D Ratio of loss of utility from loss of one year of life expectancy ÷ loss of utility from 1- θ - see Carthy et al. (1999). Hence infer the condition X is given by Response D - θ ˆ of wealth for risk of losing one year of life marginal rate of substitution, M Y
expectancy as: 1- θ ˆ = MX M Y Response D - θ . 3.3
(29)
The way forward
The most significant issue to emerge from the above discussion is the difference between linear proportionately-adjusted individual values and group aggregates. In the view of those who have prepared this protocol, there is little doubt that, while the group aggregate approach carries some conviction, the linear proportionatelyadjusted individual valuation approach is fundamentally flawed. The essential reason for this is that, with income effects and binding mental-accounting, sub-budget constraints would almost certainly entail that, while WTP would indeed be an increasing function of the size of risk-reduction/gain in life expectancy, it would also be strictly concave, rather than linearly proportionate. More specifically, in the case of Approach 1, it would appear that those who have advocated this approach have thought of the VPF upon which the approach is based as a value of (on average) 40 years of life expectancy and then have simply divided this figure by 40 (with appropriate discount factors and quality of life adjustments) to obtain the value of a life-year. To the naïve or casual reader this would then be naturally interpreted as the average sum that an individual could be expected to be willing to pay for a one-year gain in life expectancy. On careful reflection, however, this is plainly not the case and if the resultant sum is to be given any sort of credible justification it really has to be thought of as the kind of group aggregate WTP for a risk reduction that, taken across a large group of people, entails an expected gain of one year in life years lived, as under Approach 2b and Approach 3.
24
However, as noted above, Approaches 1 and 3 also rely upon another rather doubtful assumption, namely that WTP for a reduction in the risk of immediate premature death is no more and no less that WTP for the preservation of a given number of equally-valued future life years (e.g. for our 35 year old, 40 more years). However, it would not be surprising if many people’s WTP to reduce mortality risks depends on a great deal more than future life-span and could, for example, be substantially influenced by considerations such as the emotional costs of premature death to those who would be bereaved; the will to live and a concern about failing to achieve specific lifetime aspirations, such as desire to see one’s children and grandchildren grow up. In view of this, Loomes (2002) suggests that, at least beyond middle age, the VPF may be better viewed as comprising the sum of a constant component that is independent of age and reflects the pure value of living per se and a component that declined as the remaining life expectancy falls. Approach 4 involves the estimation of these parameters of such a function from existing empirical estimates of the willingness to pay for safety vs age relationship which tends to take an inverted – U life cycle form, peaking in middle age and declining thereafter. In the case of Approach 5, as noted above, the impact of income effects almost certainly means that even in the case of a one-month gain in life expectancy, WTP responses are likely to have been constrained by binding mental accounting subbudget constraints so that the first temptation is naturally to think in terms of asking about WTP for a genuinely marginal gain in life expectancy of just a few hours or days. If it were the case that respondents appreciated that such a gain could, in fact, only be realistically effected by reducing current and future hazard rates (i.e. age-conditional probabilities of premature death) then one might expect WTP responses of the same order of magnitude as those underpinning current VPF estimates. Unfortunately, however, most respondents will almost certainly think of a gain in life expectancy as an “add-on” at the end of life and so would regard a gain of just a few days in, say, 40 years time as being essentially trivial. For this reason, it was suggested that if the use of Approach 5 is to be realistically contemplated then this would require the inference of implied WTP for a few days gain in life expectancy from a WTP/gain function fitted to the actual one, three and six months WTP responses. Alternatively, the “triviality add-on” problem might be tackled by asking, not about WTP for a small extension of length of life, but rather for a small gain in quality of life, as under Approach 6. The WTP-based value of a QALY could then be estimated in one of two non-mutually exclusive ways. First, one could aggregate across the appropriate number of people to obtain total WTP for a number of small gains in the quality of life that summed to one. Alternatively, one could apply the CV/SG chained methodology detailed in Carthy et al. (1999) to obtain a marginal rate of substitution of wealth for risk of losing one year of life expectancy from a weighted-average of the responses to questions concerning WTP and willingness to accept compensation for small variations in the quality of life
25
and an appropriately-framed standard gamble (or “modified” standard gamble) question. In an ideal world the two estimates would be much the same – see for example, Bleichrodt und Quiggin (1999) – though in reality the former could be expected to be somewhat smaller than the latter.
26
4.
ESTIMATING A WTP-BASED THE USE OF DCES
MONETARY VALUE OF A
QALY
BY
Once again, for purposes of comparison, the case for using DCEs will be built on using the EQ-5D. This way, a DCE will be devised to provide a direct comparison to a WTP-based vale of a QALY derived using the contingent valuation approach outlined in the previous section. As with contingent valuation, DCEs can also be used to take into account contextual factors which emanate directly from the health care commodity in question. However, DCEs come into their own in the context of this project by allowing us to build more social factors (such as severity of illness, age and culpability) into the valuation process whilst remaining within one of the frameworks used to derive a baseline ‘generic’ monetary value of a QALY. 4.1
Defining the attributes for the baseline valuation
Given that the DCE approach is an attribute-based approach, the first stage involves breaking the valuation question down into a set of attributes. As with approaches 5 and 6 from section three, to establish a baseline value using this approach, the starting point involves three variables: EQ-5D (to reflect health-related quality of life), quantity of life (to reflect survival) and a price proxy (to allow for monetary valuation of the other two attributes)14. In defining attributes, the first stage will be to choose a reference state of health, to which no gains or losses of money are attached. This might involve, for example, describing a state of health to a 35 year old respondent which, according to the EQ-5D descriptor system, has a “utility” value of 0.9 (i.e. relatively healthy) for the remaining 40 years of their life, with the price proxy set to zero. Each respondent would be given a questionnaire involving a series of pairwise choices. Each choice would involve the reference state being compared with one of several other scenarios, in which the levels of each of the attributes in the other scenarios would vary around those specified in the initial reference state – for example, the reference state might be compared with a scenario in which an EQ-5D state equivalent to 0.95 (for the next year only, after which the respondent returns to the initial state) is presented along with no change in survival and a one-off monetary price of £500 to be paid. The choices made by respondents would reveal the trade offs between attributes, and thus the relative value to be attached to each attribute. The general method for quantifying these trade-offs, and, in this case, the monetary value of a QALY, are outlined in the following sub-section.
14
Ranges for the monetary attribute could be based on the magnitude of WTP response required to arrive at the WTP-based values for a QALY under approaches 1-4 in section three.
27
4.2
Modelling a WTP-based value of a QALY
The decision-making process within a DCE can be seen as a comparison of indirect utility functions. The subject makes a series of choices. For each choice she chooses the alternative that leads to the higher level of utility. Thus, if Uiq(A)=viq(A)+εiq
(1)
where Uiq(A) represents the indirect utility function of individual q for good i with attributes A, viq(A) represents the measurable component of utility estimated empirically, with i, q and A as defined above, and ε reflects the unobservable factors, the subject will choose i over j if: (viq+εiq) > (vjq+εjq)
(2)
or (viq- vjq) > (εjq -εiq)
(3)
Given that error terms are unknown, a probability model is estimated where: P(iqA,C)=Piq=[(εjq - εiq) < (viq – vjq)]
(4)
The probability of choosing alternative i (over j) by individual q, given the set of attributes A and the choice set C, is given by the probability that the error difference is smaller than the difference in the “observable” utility component between i and j. For purposes of empirical measurement, a probability distribution is assumed for (εjq εiq). Logit and probit techniques are commonly used to estimate the measurable component of the utility function. Given that multiple observations are obtained from individuals (since individuals are presented with a series of choices) panel data models are appropriate. When analysing the data assumptions must be made about the functional form of the observable indirect utility function. Assuming a linear utility function (as is most commonly done), then: v= α+ Σβn Xn + βn+1P + θ
(5)
where α, βn, and βn+1 are the parameters of the model to be estimated, Xn represents the levels of the n attributes of the commodity being valued (1,2…,k,…n), Σβn Xn represents the summation of all the model effect coefficients, P is the price level (or some proxy for price), and θ the unobservable error term for the model. α reflects the constant term in the model. The β parameters are equal to the marginal utilities of the given attributes (i.e. ∂v/∂Xk = βk), and the ratio of any two parameters show the marginal rates of substitution between attributes. Following on from this, the ratio of any given attribute to the absolute parameter on the price attribute shows how much money an individual is willing to pay for a unit change in that attribute (i.e. βn /βn+1).
28
Relating this to the WTP-based valuation of a QALY, and the above choice framework, the attributes would be quality of life specified from EQ-5D (QoL), length of life (L) and a price proxy P. Following the types of choices specified in section 4.1, the following equation would be estimated: V = α + (β1QoL x S)+ β2L + β3P
(6)
The coefficients from Equation (6) could then be used to estimate the rate at which individuals are willing to trade-off one attribute for another. Thus, the amount of money an individual is willing to give up for a unit improvement in QOL is equal to: (β1xS)/β3 where S is a scaling factor, based on how a unit change in quality of life is specified (i.e. whether it is to one or two decimal places) Similarly, WTP for a unit change in length of life, L is: β2/β3
Parameters β1 and β2 indicate the relative importance of QOL and L. Where preferences for one of these attributes is determined by the level of another attribute, interaction terms can be introduced into the model. Following estimation of Equation (6) it is possible to estimate WTP for movements from the reference state to some alternative situation. To do this the values of the attributes associated with the reference state are first substituted into equation (6). In this situation the price proxy will take on a value of zero. Following this, the values of the attributes associated with the alternative are substituted into equation (6). The value, or WTP, associated with the change is then subtracted from the value associated with the reference state: WTP = -(1/β3) (Vref – Valt)
Where Vref is the value associated with the reference state and Valt the value associated with the alternative state.
29
5.
GOING
BEYOND THE BASELINE ESTIMATES OF A VALUE OF A QALY
WTP-BASED
As has been argued above, it is important to extend the research to go beyond baseline WTP-based estimates of the value of a QALY in order to: • • • 5.1
assess the validity of the approach (by the use of scope tests); test for the impacts of other aspects of health care, beyond the health gains captured by QALYs, on WTP-based estimates of the value of a QALY; and take account of broader societal concerns. Testing for scope
From our review of the interventions considered by NICE, the range within which most quality-of-life improvements occur is from 0.01 to 0.3 (see Appendix B), and it is reasonable to perceive of these happening over a period of one year. Given that our basic CV/SG chained method involves a quality-of-life gain of 0.01 over a one-year period, it is important to, at least, test the sensitivity of values obtained to different quality-of-life gains over a one-year period up to the value of 0.3. Therefore, we could ask respondents to value quality-of-life changes of 0.1 and 0.25 over a one-year period. Obviously, the baseline against which values for these improvements would be assessed will be those resulting from the main survey. Thus, three extra samples would be required. The first two would separately value the 0.1 and 0.25 changes in quality of life, whilst a the third group would be asked to value both of these in addition to the initial scenario involving a 0.01 quality-of-life improvement. This combination provides a rigorous test of the sensitivity of WTP to scope (through the split sample tests) as well as providing the method with its best chance of producing results which match theoretical expectations by having a group of respondents value different sizes of health gain within the one questionnaire (a within-sample approach). A small sub-sample from the last group of respondents could also be selected for follow-up interviews in order to explore reasons for (in)consistencies found (e.g. why respondents are not willing to pay more for gains of greater magnitude). Scope tests are automatically built into the WTP-for-life-expectancy and DCE approaches, so no extra survey work would be required for these. 5.2
Testing for informational (or “process”) effects
To test the effect of a variation in information provided, three alternative versions of the contingent-valuation-based questionnaires could be presented, each to a different group of respondents. Once again, the first version would simply be that resulting from the baseline survey. In order to test for the effect of providing extra information a second group (comprising respondents to the CV/SG chained and WTP-for-lifeexpectancy approaches) would receive some additional information about the process of treatment: for example, in a recent study, respondents were informed that, “after the intervention, patients would spend three days in an intensive care unit and then 10 days in conventional hospitalisation” (Protière et al., 2003). This information was thought to be “neutral” in the sense that it simply describes what would happen, on average, to patients. It may be that the mention of intensive care would be expected to diminish WTP, as in the case of the previous study conducted by Donaldson and Shackley (1997). However, depending on prior knowledge, it may also be the case 30
that the mention of such intensive aspects of care would increase WTP. The main intention of having this second group is simply to observe the effect of providing additional information to respondents. The third group would receive not only the same “neutral” information as the second group, but also some unambiguously “positive” information about the quality of care during treatment: such as being informed that “patients could choose whether or not to be in a single room during the conventional hospitalisation period without additional cost”. To reiterate, the reason for having these three groups is to attempt to isolate the effect of providing additional (neutral) information on process from the effect of providing positive information on quality of care. It can be regarded as a way of adding an element of “control” to test whether positive information has effectively a positive impact on WTP. For the DCE method, one or more additional attributes will have to be defined to cover “process”. This is relatively straightforward, but would require yet another group to be surveyed. 5.3
Testing for impact of health gains being in terms of quality of life only, age and health status on self-only valuations
For the CV/SG chained approach, no additional data collection is required to estimate the above impacts. According to the arguments outlined on pages 21-23 above, the deviation of empirical results from the postulated equivalence of values for the same QALY gain, whether that gain is made up of quality-of-life gains only or quality and length of life gains, can be tested using the data collected in the surveys. Likewise, regression analysis can be used to test the impact of age and heath status on self-only valuations. Note, however, that such valuations are different to those of asking respondents, as third parties, to assess whether the age (and other characteristics) of recipients of care should influence the value placed on such care by “society”. This issue is dealt with in sub-section 5.4. For the WTP-for-life-expectancy approach, however, additional interviews would be required to test for the impact of expressing the same QALY gain in terms of quality and length of life improvements rather in terms of gains in years of normal health. 5.4
Using the DCE approach to investigate broader societal concerns
Our assessment is that the DCE approach is particularly suited to the task of taking account of broader societal concerns. To address this aim, respondents would be asked to place themselves in the role of a more detached citizen. Scenarios being defined in much the same way as described in sub-section 4.1 above, where we described how a DCE approach would be used to estimate a baseline WTP for a QALY. There is a general question of how far to go with this, which would be partly influenced by the complexity of the design. However, it would be relatively straightforward to extend the model to include characteristics of patients such as age, number of dependents, severity of illness, type of disease and cause/culpability, and, thus, to estimate the marginal impact on WTP values of changes in these characteristics and aspects. These marginal impacts would be estimated via the utility parameters βi.(i = age, number of dependants, cause/culpability etc). The size of the parameters would indicate the marginal impact of these factors on overall utility, and the ratio of these parameters to the price proxy the marginal WTP for a unit change in these attributes i.e. how WTP changes as age increases by 1 year.
31
6.
SURVEY METHODS
To remind the reader, the aims of the research, stated in the introductory section of this protocol, are fourfold, these being to: 1.
2.
3. 4.
6.1
obtain baseline empirical estimates of WTP-based values of a QALY, when QALY gains comprise (a) gains in life-expectancy alone and (b) life years adjusted for their quality. Given how new the proposed research is, it is, in our view, important to generate these values by more than one procedure (for “triangulation” purposes). The procedures, which have been explained above, include the following: a contingent valuation/standard gamble (CV/SG) chained approach; WTP for gains in life-expectancy in normal health (1 month, 3 months, 6 months); and discrete choice experiments (DCEs). compare these empirical estimates with various back-of-the-envelope estimates derived from existing values for prevented fatalities (VPFs) (such as the Department-for-Transport figure of £1.25 million), with these back-of-theenvelope estimates being based on refinements of the first four approaches presented in section three above. examine the manner in which other factors (e.g. “process” of care, size of health gain, age and health status) might impact on the baseline values of a QALY. extend the analysis (which so far is on a self-only basis) to accommodate wider “societal” concerns, thus attempting to value factors such as age, severity of initial health state and culpability, with the respondent taking the perspective of a more detached “citizen”. This could be done using all three approaches listed in 1., but is probably best accommodated through the use of DCEs. Using the three approaches to derive a baseline value (Aim 1)
There would appear to be two broad ways of organizing the fieldwork required to generate the individual willingness to pay/accept data that would be needed to estimate the value of a QALY. The first way of proceeding would be to conduct a specified number (e.g. 750 or 1000) of face-to-face interviews with a stratified random sample of the population, the interviews being conducted on a one-to-one basis in respondents’ homes by professional interviewers from a reputable sample survey organization. Given the nature of the questions that it would be necessary to put to respondents, these interviews would be most efficiently carried out using lap-top computers with a specially prepared interactive programme as was the case, for example, in a national value of road safety survey carried out in New Zealand in 1997/98 – see Guria et al. (1999) – or in a more recent DEFRA-funded project on valuing reductions in air pollution. While it would be a straightforward matter to conduct in the region of 1000 interviews, the most significant disadvantage of this approach is that although the interviewers would be professionally-trained, experienced and – one might reasonably expect – competent, it would be unreasonable to expect them to have the economic/cognitive psychological expertise necessary to deal with some of the sorts 32
of questions that would almost certainly be posed by at least a subset of respondents (e.g. “why are you asking questions like this?” or “I know what’s going on here – this is the thin end of the wedge for tax increases to fund the NHS” etc. etc.). By contrast, the second possible approach would be based on focus group sessions involving between four and six randomly selected participants - moderated by members of the research team and research associates - using a pre-prepared protocol and involving a combination of open-ended group discussion of pre-determined questions and issues, together with the completion of individual questionnaires by members of the focus group.15 For example, the group discussion would be divided into a number of phases. Phase 1 (Introduction) would provide an overview of the objectives of the group/project, giving participants the framework within which they would be working. Participants would be informed about the role of the project sponsors and why they had commissioned this project in very broad terms (this reduces any initial, and understandable, initial suspicions they might have). This would be followed by any key information necessary for the discussion to commence. In the case of the CV/SG chained groups, participants would be introduced to the concepts of WTP and willingness-to-accept via a number of warm–up exercises prior to answering such question “for real”. This Phase could, if desired, be followed by a qualitative discussion around their answers. The next two Phases would be given over to eliciting the SG responses (again, through a mix of warm-up, quantitative and qualitative exercises). More open-ended discussion can be inserted at appropriate instances within the procedure if (i) more general information/opinions are required about these (or other) issues and/or (ii) the respondents require a break from the quantitative tasks, either because of fatigue or automatum. The final Phase concludes with a general debriefing session. The clear advantage of this type of procedure is that the focus group moderators would have the expertise necessary to give them a reasonable chance of fielding awkward questions of type outlined above. On the other hand, given that focus groups would realistically require at least two moderators per group, this procedure is clearly somewhat more labour-intensive than the previous one and since the team from which focus group moderators could be selected is necessarily limited in size, it would be unrealistic to expect the procedure to generate a sample of more than, say, 200-250 15
Readers will have noticed that little has been said about the exact details of the questionnaire, although the content of key questions has been outlined in earlier sections. It is worth noting here that another important issue which may be thought worthy of further investigation is the payment vehicle to be used. However, we do not see this issue as being that important in this particular study. We propose to use a payment-scale approach because: (a) although the closed-ended (or referendum) approach was recommended in the landmark NOAA-Panel guidelines for the conduct of WTP studies (Arrow et al., 1993) and has demonstrated validity in health care vis-à-vis open-ended questions (Johannesson, 1992; Johannesson et al., 1991, 1993), recent research demonstrating the potential for ‘yea-saying’ has cast doubt on its superiority over other vehicles (Holmes and Kramer, 1995; Kramer and Mercer, 1997; Ready et al., 1998); (b) the payment scale has also been shown to be valid relative to open-ended questions in the health context (Donaldson et al., 1998); and, (c) the research team has substantial experience with the use of the payment-scale method in both health and safety projects. Other researchers, bidding to conduct a survey to elicit empirical estimates of the WTP-based value of a QALY could propose other (valid) methods with which they feel comfortable, bearing in mind that the closed-ended approach would also require larger samples due to the nature of information collected.
33
respondents within a realistic time period. A further potential advantage of this approach, however, is that it would be capable of providing fairly substantial qualitative insights as well as generating the required quantitative data, as the focusgroup method eventually works round to respondents each completing a questionnaire for quantitative analysis. The opinion of those who have prepared this protocol is that while either of the two above procedures could probably be applied to approach 5 detailed in section three and to DCEs, it is probably unrealistic to expect that one could proceed with other than the focus-group procedure under approach 6 from section three (i.e. the CV/SG chained approach). However, we again regard relative advantages and disadvantages of the two procedures as being essentially a matter for those who respond to the protocol to evaluate. Given that to conduct interviews with 750-1000 people for each approach (i.e. 2250-3000 in total) would likely be financially prohibitive as well, we have budgeted below for each approach to be applied to separate samples of 250 respondents (i.e. 750 in total). We feel that a word on qualitative analysis is merited at this juncture. The aim of the qualitative data collection will be to an extent dictated by the specific survey and issue under investigation, but, generally speaking, it is used within the focus group process to identify potential knowledge, definitions, heuristics and misunderstandings that respondents may bring with them to the valuation exercise. Whilst potentially a rich source of data, something more than an unstructured “fishing expedition” is required to ensure the researchers do not “hear what they want to hear” from such groups, thereby introducing unintended errors into the analysis of which they are not conscious. While there are a number of qualitative methods available, Chilton and Hutchinson (1999) identify two in particular that are potentially suitable for the type of focus group discussions under consideration – grounded theory (Glaser and Strauss, 1967; Turner, 1981; Henwood and Pidgeon, 1992) and content analysis (Holsti, 1969; Krippendorf, 1980). The ultimate choice regarding the most appropriate method (or methods) will be ultimately be one for the successful team, conditional on the type of questions/discussions they intend to use. This, we only briefly outline these techniques at this stage. Grounded theory does not impose a predetermined framework on the discussion. The analyst examines the transcript of the discussion, beginning with a few very general “priors”, building up a list of codes and theories from the discussion, recording and amending them as they go along in the light of new information. This method allows the data to “speak” to the analyst with as little mediation as possible from any a priori theoretical prejudices. It has the potential to provide very detailed knowledge about an issue, although it is very labour intensive. Content analysis imposes a predetermined framework on the discussion, although this can be changed as the study progresses and/or the issues to be investigated change. This is a robust form of analysis which allows qualitative factors to be “quantified” and counted in a reliable and semi-objective manner. The aim of this technique is not to capture all available information, rather it is to capture, in a reliable manner, the key points from the data. Using content analysis, it is possible to isolate concepts from the data and to assess their importance in generating the observed valuations. Applying this to the subjects’ written responses provides a set of codes which can be
34
used to test various hypotheses relating to the strategies utilized by them in determining their two values. 6.2
Comparisons of CV/SG chained, WTP for gains in life expectancy and DCE-based values with existing values of safety (Aim 2)
As far as aim 2 is concerned, there are at least two ways in which one might assess the plausibility of an estimated WTP-based monetary value of a QALY in relation to other existing WTP-based values such as those used in the transport safety context. First, as outlined in section three, it is a relatively straightforward matter to convert an estimated value for a QALY into a rough and ready equivalent value for the prevention of a statistical fatality simply by regarding the latter as remaining life expectancy times the value of a QALY (possibly with some adjustment for discounting). Thus, suppose that the value of a QALY is estimated as being £30,000. Over remaining life expectancy of, say, 40 years this would then without discounting sum to a total VPF of £1.2 million, which is very close to the current Department for Transport VPF. Alternatively, the estimated monetary value of a QALY could be compared more directly with the Department for Transport WTP-based monetary value for the prevention of the appropriate non-fatal road injury. Again as an example, the study on which the current Department for Transport non-fatal injury values was based – see Jones-Lee et al. (1995) – included a variety of severities of non-fatal road injury, one or some combination of which would almost certainly be equivalent to the loss of a QALY or some multiple of the latter close to unity and against which the estimated value of a QALY could be directly compared. 6.3
Investigating other factors (Aim 3)
All of the surveys detailed in sub-sections 6.3 and 6.4 will be based on the same methods (i.e. based on the use of focus groups) as outlined above. Therefore, in the main, we concentrate on the numbers required to undertake the experiments outlined in section five. 6.3.1
Testing for scope
As explained above, scope tests are built into the WTP-for-life-expectancy and DCE approaches. For the CV/SG approach scope tests would involve three samples, in addition to the group used to estimate the baseline WTP-based value of a QALY. The first two groups, of 100 each, will separately value 0.1 and 0.25 changes in quality of life (as opposed to 0.01 in the baseline survey). The third group (n=100) will be asked to value both of these in addition to the initial scenario involving a 0.01 quality-of-life improvement. As we have said above, this combination provides a rigorous test of the sensitivity of WTP to scope (through the split sample tests) as well as providing the method with its best chance of producing results which match theoretical expectations by having a group of respondents value different sizes of health gain within the one questionnaire (a within-sample approach).
35
A sub-sample of 30 of the last group of respondents will be selected for follow-up interviews in order to explore reasons for (in)consistencies found. Thus, the total number of surveys required for scope testing is 330. 6.3.2
Testing for informational (or “process”) effects
Once again, one group will be formed by the initial survey to estimate the baseline WTP-based value of a QALY. To test for the effect of providing extra information a second group (n=75, for each of the CV/SG chained and WTP-for-life-expectancy approaches, or 150 in total) will receive some additional information about the process of treatment. A third group (n=150) will receive not only the same “neutral” information as the second group, but also some unambiguously “positive” information about the quality of care during treatment For the DCE method, one or more additional attributes will have to be defined to cover “process”. This is relatively straightforward, but would result in an additional 100 questionnaires having to be administered in order to investigate this issue through the DCE method. Therefore, to assess for informational effects, will require an additional 400 surveys. 6.3.3
Testing for impact of health gains being in terms of quality of life only, age and health status on self-only valuations
As stated in sub-section 5.3, no additional data are required for CV/SG chained and DCE approaches. For the WTP-for-expectancy approach, however, an additional 100 interviews would be required to test for the impact of expressing the same QALY gain in terms of quality and length of life improvements rather in terms of gains in years of normal health. 6.4
Using the DCE approach to investigate broader societal concerns (Aim 4)
To address this aim, respondents will be asked to place themselves in the role of a more detached citizen. Our estimate is that, taking account of such factors would result in approximately 200 more questionnaires being administered.
36
6.5
Timetable and budget
6.5.1 Timetable
Overall, we would estimate the project to last two and a half years, in four main stages, as outlined below: Number of interviews
Months
Activity
1-9
Questionnaire design Scenario development Design of computer programmes for questionnaires Piloting
80
10-18
Fieldwork
750
19-21
Analysis/reporting of baseline estimates
22-30
Fieldwork and analysis for work on valuing other aspects of health care and societal concerns
6.5.2
1030
Budget
Over the duration of the project, three of the academic staff, who each have full teaching loads, would have their time bought out to participate in the research project, much of this involvement taking up private time (e.g. for focus groups in evenings). The grades involved are a professor, senior lecturer and lecturer, each inputting one and a half days per month (for 30 months) into questionnaire design, scenario development (which will require much refinement beyond what is presented in this protocol), piloting, fieldwork, analysis and writing at rates of £607, £441, and £365 (inclusive of overhead) respectively. This amounts to £63,585 (£44,510 for the baseline estimates and £19,075 for the subsequent work on other aspects and societal concerns). During months 1-9, we will require 3 months of a computer technician’s time in order to prepare questionnaires in computerised forms. This task also requires development of an interactive programme which will allow the computer to calculate (based on respondent’s self-assessed health) a given percentage reduction in quality of life and, hence, a correspondingly worse health state to present to the respondent for valuation. The cost of this (including overhead) will be £10,886. We propose to employ one and a half full-time equivalent research associates/fellows to support all the activities associated with the administration (i.e. moderation) of the CV/SG chained, WTP-for-life-expectancy and DCE approaches for the full 30 months of the project. These people will be responsible for project co-ordination, the main task here being the moderation of the focus groups for the three sets of parallel surveys to produce baseline estimates of WTP for a QALY. They will assist with all other aspects of the project, these being questionnaire design, scenario design, piloting, fieldwork, analysis and writing. Total cost for this is £169,376 (£118,563 for the
37
baselines estimates and £50,813 for the subsequent work on other aspects and societal concerns). This cost is inclusive of overheads for a person on research scale 1A, scale point 10. Casual assistance is also required for moderation of focus groups and for transcription of material, as these are onerous tasks. Whilst some of these tasks will be performed by the research associates and the academic staff involved in the project, we require 600 hours at £20 an hour for this task, costing £5000 for production of baseline estimates and £7000 for other aspects and social concerns; £12,000 in total. We would propose to recruit groups across four sites in England and Wales. Thus, costs of about £130 per interview will be incurred for recruitment, room hire, payments to participants, reimbursement of participants and travel and subsistence for researchers to moderate the groups. This will result in costs of £97,500 for piloting and baseline estimates and £133,900 for the subsequent work on other aspects and societal concerns, a total of £231,400. Computers are required for each of the research associates, a cost (of £5,000) which has been allocated to the baseline survey costs. Sundry materials, such as those for survey production, have been costed at £5 per interview – a total of £9,300, or £4150 for pilot and baseline surveys and £5150 for the other aspects and societal concerns. Total costs for the two stages of the project combined come to £501,547. For each stage, the costs are summarised in the following table: Summary of costs Item Academic inputs Computer technician Research associate/fellow Moderating assistance Interview costs Computers Sundry materials
Piloting and surveys £44,510 £10,886 £118,563 £5,000 £97,500 £5,000 £4150
Totals
£285,609
baseline Other aspects/societal concerns £19,075
£50,813 £7,000 £133,900 £5150 £215,938
38
APPENDIX A
WILLINGNESS-TO-PAY-BASED VALUES OF SAFETY IN PUBLIC SECTOR PROJECT APPRAISAL
39
Of great relevance to the issue addressed in the proposed research has been the work related to the above issues within the context of valuation of safety policies. This work has involved the development of methods to elicit a ‘value of statistical life’ and, thus, is more closely related to the challenge of valuing a QALY (or ‘healthy year’). Given its subject matter, it is also one of the most controversial issues in the appraisal of proposed public sector projects. Until the 1980s most countries that explicitly addressed the public sector safetyvaluation issue tended to use some variant of the so-called “gross output” or “human capital” approach. Under this approach the primary component of the “cost” of the premature death of an individual is treated as the discounted present value of that individual’s future output extinguished as a result of his or her premature demise. In some countries (including the UK) a further more-or-less arbitrary allowance was then added to the gross output figure to reflect the “pain, grief and suffering” of the victim and/or his/her surviving dependents and relatives. Values for the prevention of premature death are then defined in terms of the costs avoided. To give an example of the costs and values that emerge under the gross output approach, the UK Department for Transport’s most recent gross output – based value for the prevention of a road fatality – based on national averages – was £180,330 in 1985 prices, of which about 28% was an allowance for pain grief and suffering. Updated for inflation and growth of real output per capita this figure would now stand at some £460,000 in 2002 prices. Not surprisingly, many economists have objected to the gross output approach on the grounds that most people almost certainly value safety largely because of their aversion to the prospect of their own and others’ death and injury as such, rather than because of a concern to preserve current and future levels of output and income (Schelling, 1968; Mishan, 1971; Jones-Lee, 1989). Given this, it has been argued that values of safety ought ideally to be defined so as to reflect people’s “pure” preferences for safety, per se, rather than in terms of effects on output and income, as in the gross output approach. However, in order to define and estimate values of safety in this way we clearly require some means of measuring people’s preferences for safety and, more particularly, their strength of preference. How can one do this? Arguably, the most natural measure of the extent of a person’s preference for anything is the maximum amount that he or she would be willing to pay for it. This amount reflects not only the person’s valuation of the desired good or service relative to other potential objects of expenditure, but also the individual’s ability to pay – which is itself a manifestation of society’s overall resource constraint. So, under what has naturally come to be known as the “willingness-to-pay” (WTP) approach to the valuation of safety, one first seeks to establish the maximum amounts that those affected would individually be willing to pay for (typically small) improvements in their own and others’ safety. These amounts are then simply aggregated across all individuals to arrive at an overall value for the safety improvement concerned. The resultant figure is thus a clear reflection of what the safety improvement is “worth” to the affected group, relative to the alternative ways in which each individual might have spent his or her limited income. Furthermore, defining values of safety in this way effectively “mimics” the operation of market forces – in circumstances in which markets typically do not exist – insofar as such
40
forces can be seen as vehicles for allowing individual preferences to interact with relative scarcities and production possibilities to determine the allocation of a society’s scarce resources. In order to standardise values of safety that are derived from the WTP approach and render them comparable with values obtained under other approaches (such as gross output), the concept of the prevention of a “statistical” fatality or injury is applied. To illustrate this concept, suppose that a group of 100,000 people enjoy a safety improvement that reduces the probability of premature death during a forthcoming period by, on average, 1 in 100,000 for each and every member of the group. The expected number of fatalities within the group during the forthcoming period will thus be reduced by precisely one and the safety improvement is therefore described as involving the prevention of one “statistical” fatality. Now suppose that individuals within this group are, on average, each willing to pay $w for the 1 in 100,000 reduction in the probability of death afforded by the safety improvement. Aggregate WTP will then be given by $w x 100,000. This figure is naturally referred to as the WTP-based value of preventing one statistical fatality (VPF) or alternatively as the value of statistical life (VOSL). Clearly, in the above example, average individual WTP, $w, for the average individual risk reduction of 1 in 100,000 is a reflection of the rate at which people in the group are willing to trade off wealth against risk “at the margin”, in the sense that the trade-offs typically involve small variations in wealth and small variations in risk. Empirical work on the valuation of safety thus tends to focus upon these individual marginal wealth/risk trade-off rates. On a somewhat more cautionary note, it is extremely important to appreciate that, defined in this way, the VPF is not a “value (or price) of life” in the sense of a sum that any given individual would accept in compensation for the certainty of his or her own death – for most of us, no finite sum would suffice for this purpose, so that in this sense life is literally priceless. Rather, the VPF is aggregate WTP for typically very small reductions in individual risk of death (which, realistically, is what most safety improvements really offer at the individual level). Similarly, in the case of the WTPbased monetary value of a QALY, as argued in section 3.3, this is probably also most appropriately viewed as a group-aggregate willingness to pay for marginal gains in quality of life or life expectancy given that, at least in the case of a randomly-selected sample of the public, such gains will typically be marginal, though of course the same cannot necessarily be said for those already suffering from health impairments. Before outlining the various ways in which researchers have sought to obtain empirical estimates of values of safety using the WTP approach, two further points should be noted. First, so far only passing reference has been made to people’s concern – and hence WTP – for others’, as well as their own safety. Insofar as people do display such “altruistic” concern then one would naturally expect that it would be appropriate to augment the WTP-based VPF to reflect the amounts that people would be willing to pay for an improvement in others’ safety. However, it turns out that under plausible assumptions about the nature of people’s altruistic concern for others’ safety on the one hand and their material wellbeing on the other (the latter being reflected by their wealth or consumption), augmenting the VPF to reflect WTP for other’s safety would involve a form of double-counting and would therefore
41
ultimately be unjustified. For example, suppose that individual A is concerned not only about individual B’s safety, but also about the latter’s wealth or consumption. Furthermore, suppose that individual A’s altruistic concern for B is “pure”, in the sense that it respects B’s preferences. While A will then regard a reduction in B’s risk of premature death as a “good thing”, she will also regard the increase in B’s taxation (or other expenditure) required to finance the risk reduction as an exactly offsetting “bad thing”. Taking account only of A’s WTP for B’s safety improvement would therefore quite literally involve double-counting. Thus, the issue of whether and how peoples’ concern for others’ safety ought to be taken into account under the WTP approach hinges on the essentially empirical question of the relationship between such concern and concern for others’ wealth or consumption. For a more detailed discussion of the issue of altruism and safety, see Jones-Lee (1992). Given all of this, it would seem that similar arguments apply to the WTP-based monetary value of a QALY. A further important aspect of the WTP approach involves recognition of the fact that safety improvements also have “direct” economic effects, such as avoidance of net output losses (i.e. losses of the excess of an accident victim’s future output over his/her future consumption) material damage, medical and police costs and so on. To the extent that people appear in the main not to take account of such factors in assessing their WTP for improved safety (and there is some evidence that they tend not to – see Jones-Lee et al. (1985)) then an allowance for these factors should clearly be added to WTP-based values of safety. However, such additions tend to be relatively modest in relation to the typical magnitude of aggregate WTP for safety per se, at least in the case of risks of premature death. But how, in fact, are WTP values of safety estimated in practice? Broadly speaking, three variants of empirical estimation procedure have been employed to derive WTPbased values of safety. These are known respectively as the “revealed preference” (or “implied value”), the “contingent valuation” (or “expressed value”) and “relative valuation” approaches. Basically, the revealed preference approach involves the identification of situations in which people actually do trade off income or wealth against physical risk – for example, in labour markets where riskier jobs can be expected to command clearly identifiable wage premia (Smith, 1983; Viscusi and Moore, 1989). By contrast, the contingent valuation approach involves asking a representative sample of people more or less directly about their individual WTP for improved safety, (or, sometimes, their willingness to accept compensation for increased risk). The difficulty with the revealed preference approach when applied to labour market data is that it depends on being able to disentangle risk-related wage differentials from the many other factors that enter into the determination of wage rates. The approach also presupposes that workers are well-informed about the risks that they actually face in the workplace. In addition, those whose jobs do carry clearly identifiable wage premia for risk may not be representative of the work force as a whole, in that such people almost certainly have a below-average degree of risk-aversion (Gegax et al., 1991).
42
The great advantage of the contingent valuation approach is that it allows the researcher to go directly and unambiguously to the relevant wealth/risk trade-off – at least, in principle. On the other hand, the contingent valuation approach has the disadvantage of relying upon the assumption that people are able to give considered, accurate and unbiased answers to hypothetical questions about typically small changes in already very small risks. By contrast, unlike the revealed preference and contingent valuation approaches, the relative valuation approach does not involve an attempt to estimate wealth/risk tradeoff rates directly, but rather seeks to determine the value of preventing one kind of physical harm relative to another. Thus, for example, the UK Department for Transport’s current monetary values for the prevention of non-fatal road injuries of various levels of severity were obtained by applying estimates of such relative valuations to an absolute monetary “peg” in the form of the Department’s existing WTP-based roads VPF. Turning to the question of the figures that are actually applied in practice, WTP-based values of safety are currently used in road project appraisal in the UK, USA, Canada, Sweden and New Zealand, with several other countries employing values that have been substantially influenced by the results of WTP studies. More specifically, in the UK the Department for Transport (DfT) currently employs a figure of £1.19 million in June 2001 prices for the prevention of a statistical fatality in its roads project appraisal. This figure was based on the findings of a study which obtained estimates of the roads VPF using a variant of the contingent valuation approach – see Carthy et al (1999). In turn, the Department’s values for the prevention of serious and slight non-fatal injuries are £134,190 and £10,350 respectively, again in 2001 prices, these figures having been obtained using the relative valuation approach to estimate non-fatal/fatal valuation relativities – see Jones-Lee et al (1995), and Department for Transport (2002). In the USA, the US Department of Transportation currently values the prevention of a statistical road fatality at $US 3 million, this being an update of a figure originally recommended in 1991 following a survey of the then - existing literature on empirical estimation of WTP-based values of safety – see The Urban Institute (1991). In turn, Transport Canada applies a WTP-based value for the prevention of a statistical fatality of $Cdn1.5 million in 1991 prices based on a survey of the literature. Updated for inflation and growth this would be very close to the current DfT UK value. Finally, the WTP values used in Sweden and New Zealand were derived under the contingent valuation approach and in 1999 prices are SEK 14.30 million (roughly £1.07 million) and $NZ 2.5 million (roughly £820,000), though in the latter case it should be noted that the New Zealand Land Transport Safety Authority is considering increasing the figure to $NZ 4 million (roughly £1.32 million) on the basis of recommendations following an extensive contingent valuation study carried out in New Zealand in 1997/98 – see Guria et al. (1999). Recently, both quantitative and qualitative research has cast doubt on the reliability and validity of WTP values for safety derived through the above direct contingent
43
valuation method. As well as sequencing and framing effects, a prominent issue has been the lack of ability of the method to account for embedding and scope. That is, respondents tend to view safety improvements as a ‘good thing’ and, therefore, will often state much the same WTP for different sizes of risk reduction, whether for fatal or non-fatal injuries (Jones-Lee et al., 1995; Dubourg et al., 1997; Beattie et al., 1998). It may be unreasonable to expect respondents to give accurate answers to hypothetical questions which involve direct trade-offs between wealth and small reductions in risk. Therefore, Carthy et al. (1999) have suggested a less-direct CV/SG ‘chained’ approach which breaks down the valuation process into a series of manageable steps which involves chaining together responses to WTP and SG questions. First, respondents are presented with a question asking them about their WTP for the certainty of a complete cure for a given non-fatal road injury and their willingness to accept compensation for the certainty of remaining in the impaired health state (the combination of which, based on some reasonable assumptions about underlying preferences obeying minimal conditions of consistency and regularity, it is argued, gives a reasonable estimate of the marginal rate of substitution of wealth for the risk of the non-fatal injury). Second, respondents are presented with a SG question aimed at determining the ratio of the health state value for death over that for the non-fatal injury. The monetary value from the first stage can then be combined with the ratio from the second stage to obtain a WTP for reduced risk of death. The approach is, perhaps, more realistic in that most people can relate to giving a monetary value for avoiding a non-fatal injury of the sort they are likely to have experienced, and people are not asked directly to place a monetary value on a small risk reduction. The method has shown promise in terms of being subject to less marked embedding effects and other biases than earlier approaches (Beattie et al., 1998).
44
APPENDIX B
SUMMARY OF QUALITY-OF-LIFE ESTIMATES USED IN NICE HEALTH TECHNOLOGY APPRAISALS
45
A review of the NICE appraisals, which are published as HTA reports and available from the HTA website (http://www.hta.nhsweb.nhs.uk), produced the following information about quality of life and QALY values. The reports represent a range of systematic reviews. Some present reviews of the existing literature with no additional analysis. Others synthesise the data from the literature and industry submissions and have adapted existing models or developed new models. Many of the reviews comment on the lack of good quality QOL data available. Very few could rely on utility values elicited from patients or the general public. Utility values have been estimated, using a variety of methods, from the available QOL information, e.g. disease specific measures. In summary:
53 NICE appraisals are listed as complete on the NICE website, of which 40 have corresponding HTA reports 17 reports did not give utility values 15 reports estimated QALYs but QOL gains/ survival cannot be extracted 8 reports include estimates of 20 QOL gains in the range 0.0110.461, mean 0.19, median 0.15 18 of the 20 values are 0.284 and below
NICE appraisal No. 33
Title
Publication date
HTA website
QALY information. Quality of life changes.
Advanced Colorectal Cancer irinotecan, oxaliplatin & raltitrexed (NO 33)
January 2002
Vol 5 25
19
Alzheimer's disease donepezil, rivastigmine and galantamine (N0 19)
January 2001
Vol 5 1
Utility values are not measured in any of the trials. QOL is mainly measured by cancer QOL life scale (EORTC) and differences in QOL are insignificant or ambiguous in several trials. Cost effectiveness usually reported as cost per progression free year. These values are then adjusted for QOL using utility values from literature (nurse proxies) for colorectal cancer states of 100, 95, 57.5, 10, for progressively worse disease states from partial response to terminal disease. Authors state that utility examples are illustrative only and conclusions should not be drawn from them For 1st line treatment, QOL adjustment reduces the estimate of progression-free survival by 5.4%-6.5% for oxaplatin and 10.7-19.2% for irinotecan. Progression free survival benefit from 1st line treatments ranges from 1.69 – 2.77 months. Cost per QALY not attempted – lack of info
46
11
Arrhythmias - implantable cardioverter defibrillators (N0 11)
September 2000
Vol 4 26
38
Asthma - inhaler devices for older children (No. 38)
April 2002
Vol 6 5
10
Asthma - inhalers for children under five (N0 10)
August 2000
13
Attention deficit hyperactivity disorder (ADHD) - methylphenidate (N0 13) Brain cancer - temozolomide (N0 23)
October 2000
Not listed on HTA pages Not listed on HTA pages 5(13)
6
Breast cancer - taxanes (N0 6)
June 2000
30
Breast cancer - taxanes review (N0 30)
September 2001
Not listed on HTA pages * 4(17)
34
Breast cancer - trastuzumab (NO 34) Cervical smear tests - liquid based cytology (N0 5) Colorectal cancer laparoscopic surgery (N0 17)
March 2002
6 (13)
June 2000
4(18)
December 2000
Not listed on HTA pages Not listed on HTA pages
23
5 17
40
Crohn's disease (No. 40)
April 2001
March 2002
No published QOL data. Utility gain (based on expert opinion and HRQOL scale) estimated at 0.08. Survival curve analysis indicates 20 additional life years gained for every 100 patients treated for 3 years. Estimated cost per life year saved is £40,500-£87,000 Estimate 0.38 QALYs gained over 3 years with ICD over drug therapy Additional costs of £8100-£17400 Cost per QALY gained with ICD £21300-£45800. Authors warn that this is speculative and other studies show no gain Low mortality rate, cost per QALY not calculated.
Speculative C-U model Evidence for survival inconclusive. Utility values estimated from global QOL question in cancer QOL instrument EORTC QLQ-C30 Baseline case: increase in progression free survival 4 weeks Increase in utility 0.2 Cost per QALY gained £42920 3 scenarios assume 0, 0.2 and 0.4 increase in utility (0.4 representing return to perfect health) – resulting cost per QALY gained £24,454; £42,920; £175,256 * ? included in the ovarian cancer/ taxanes review poor QOL data – authors comment that few studies break down the QALY value Clinical effectiveness only Insufficient QOL data to estimate QALY
47
51
Depression and anxiety computerised cognitive behaviour therapy (No. 51)
October 2002
6 (22)
Utility values taken from 2 studies estimating scores for mild, moderate, severe depression, and remission from depression. Authors state that assumptions mean that estimates should be treated with caution. Cost per QALY figures related to 6 months following treatment. In the first example these are given as: 0.59, 0.32, 0.09 and 0.79 Estimated mean utility gains 0.27, 0.47 for TAU and BTB respectively over 6 months – multiplied by 0.5 for QALYs 0.135, 0.235. Incremental cost per QALY gained for BTB over TAU: £1209 - £7692
21
Diabetes (type 2) pioglitazone (N0 21) Diabetes (type 2) rosiglitazone (N0 9)
March 2001
5 (19)
August 2000
53
Diabetes - Long acting insulin analogues (No. 53)
December 2002
7
Dyspepsia - proton pump inhibitors (N0 7)
July 2000
Not listed on HTA pages Not listed on HTA pages Not listed on HTA pages
9
In the second example, estimates are taken from SG values but included drug side effects. For 3 drugs, mean utility gains for TAU and BTB respectively and incremental QALY gains for BT over TAU are given as Drug 1: mean utility gains: 0.10, 0.20; incremental QALY gain: 0.05 Drug 2: mean utility gains: 0.07, 0.17; incremental QALY gain: 0.05 Drug 3: mean utility gains: 0.11, 0.17; incremental QALY gain: 0.03 Incremental cost per QALY gained £3000 - £6667 No relevant info
48
15
Flu - zanamivir (Relenza) (N0 15)
November 2000
6 (9)
Previous study QOL data based on assumptions with utility weight for a day deduced from Quality of Wellbeing scale with additional tariff for symptoms giving score of 0.5579 Health state free from influenza = 1.0 Drug cost of £24 per day – 5 day course Cost per QALY = £6250 This report develops a new model to take into account some of the shortcomings of the existing model. Survival data are rarely reported because mortality is low. Model estimates death rate from influenza in over 65 years at 28/100,000 without zanamivir and 26.3/ 100,000 with zanamivir. Authors report no published utility scores for flu – they assume EuroQol score of 22222 = 0.516 Health state free from influenza = 0.8 Cost per QALY = £65000 all adults (circulating influenza), £54000 for ‘at-risk’ (circulating influenza) and £158,000 for all adults (influenza season). (highly sensitive to a range of parameters .)
12
Glycoprotein IIb/IIIa inhibitor guidance for acute coronary syndromes (N0 12)
September 2000
4 (30)
47
Glycoprotein IIb/IIIa inhibitor guidance for acute coronary syndromes - review (NO 47)
September 2002
6 (25)
48
Haemodialysis - home versus hospital (NO 48)
September 2002
8
Hearing disability - new advances in hearing aid technology (N0 8) Heart disease (ischaemic) coronary artery stents (N0 4)
July 2000
Not listed on HTA pages 4 (4)
May 2000
4 (23)
Hepatitis C - alpha interferon and ribavarin (N0 14)
October 2000
6 (31)
4
14
Many different studies Only 1 prospective economic evaluation (USA) Found $16,491 per life year gained or $19,693 per QALY or $23,449 using a ‘more conservative QALY rating scale outcome’ No QOL information Same literature used as above study, plus 2 industry submissions No QALY information Model developed by authors but not reported here
No relevant info No direct cost utility estimate or model Cost per QALY is reported, but no info breaking down the QALY Ref Cohen et al 1994 for QOL data Previous model is adapted for up to date costs Utility values are not stated Authors cite existing model for values
49
18
Hernia (inguinal) laparoscopic surgery (N0 18)
January 2001
44
Hip Resurfacing - metal on metal (No. 44)
June 2002
Not listed on HTA pages 6 (15)
2
Hips - prostheses for primary total hip replacement (N0 2)
March 2000
2 (20)
42
Human Growth Hormone in Children (N0 42) Juvenile idiopathic arthritis etanercept (No. 35) Knee joints (defective) autologous cartilage transplantation (N0 16) Leukaemia (chronic myeloid) - imatinib (N0 50)
May 2002
6 (18)
March 2002
6 (17)
December 2000
5 (11)
October 2002
Leukaemia (lymphocytic) fludarabine (N0 29)
September 2001
Not listed on HTA pages 6 (2)
35 16 50
29
Markov model over 20 years QOL in model based on assumptions about pain associated with different treatments and published QOL scores for mild, mod and severe OA (For THR these were 0.82,0.52,0.18 respectively), and the percentage of patients in those states, giving QOL values: MOM 0.964, THR 0.964, Watch wait 0.503, osteot 0.946, arthro 0.964 QALYs gained by MOM versus W/THR, THR, osteot and arthro: 3.73, -0.02, 1.07, 6.04 respectively Incremental Cost per QALY for MOM versus W/THR, THR, osteot and arthro: MOM dominates, THR dominates, £3,039, £366. Sensitivity analysis reported, effect on different age groups. Markov model over 60 years. assume mild moderate and severe pain utility values are 0.69 0.38 and 0.19 and no pain = 1 after op 80% patients no pain, 20% mild pain for op fails 15% patients severe pain, 85% moderate pain (benefits discounted at 6%) Baseline life expectancy, all patients - 9.19 years Expected QALYs, all patients 8.39 Threshold analysis for cost neutrality, £6500 per QALY, and £10000 per QALY gained, sub group analysis and sensitivity analysis reported QALYs not calculated – insufficient info in lit QALYs not calculated – insufficient info in lit QALYs not calculated – insufficient info in lit
QALYs not calculated – insufficient info in lit
50
26
Lung cancer - docetaxel, paclitaxel, gemcitabine and vinorelbine (N0 26)
June 2001
5 (32)
37
Lymphoma (follicular nonHodgkin's) - rituximab (no. 37) Motor neurone disease riluzole (N0 20)
March 2002
6(3)
January 2001
5(2)
20
QALYs not calculated. Most of the reviewed literature use cost per tumour response/ cost per LYG. 2 studies reporting proxy utility values from oncology professionals – giving values of a) 0.7 for VNB and 0.6 for CDDP regimes and b) 0.53 baseline, compared with 0.6 for both VNB and VNB/CDDP, 0.65 for GEM, and 0.63 for PAX/CDDP incremental median survival for all treatments range = 0.70-4.76 months incremental cost per LYS (vs BSC) range £2,194 – 46,610 QALYs not calculated – insufficient info in lit No C-U studies in the literature. One CU - NICE submission from industry - reporting £12,384 per QALY gained. (base case survival result not stated) This model uses utility scores for based on SG values for 4 health states which are progressively more severe – 0.79, 0.67, 0.71, 0.45. Discounted survival (months) with riluzole: 20.85 Discounted survival (months) with placebo: 19.24 Base case results : Life years gained 0.13, QALYs gained 0.09, increase in costs £5200 Cost per QALY £58,000 Cost per life year £39,000. Sensitivity analysis range £20,000 per QALY to riluzole dominated by placebo.
51
32
Multiple Sclerosis - beta interferon and glatiramer acetate (N0 32)
November 2001
Not listed on HTA pages *
52
Myocardial Infarction - early thrombolysis treatment (No.52)
October 2002
46
Obesity (morbid) - surgery (No.46)
July 2002
Not listed on HTA pages 6(12)
* One rapid review of 9 different treatments appears on HTA pages: 4(9) Data has been extracted from this report as follows: azathioprine: no CEA studies found beta interferon in RRMS: 3 CUA studies costs in 3 studies are $17,000; £9500 and £10500 changes in QOL are 0.018 QALYs per relapse; 0.0112 QALYs p.r. and 0.0417 QALYS p.r. best estimates for cost per QALY for 2 UK studies are £94,000 and £74,500. beta interferon in SPMS: 2 CUA studies annual costs per patient £9800 and £9600 QOL changes 0.239 QALYs gained by delays to progression; 0.281 QALYs per 9 months of wheelchair dependence avoided Costs per QALY estimated at £874,600 and £1,024,000 cladribine: No CEA studies found cyclophosamide: No CEA studies found glatimarer: one CUA study annual cost of treatment £10,100 QOL changes 0.011 QALYs per average relapse Cost per QALY approximately £500,000 (best estimate of £90,000) intravenous Ig: no CEA studies found methotrexate: no CEA studies found mitoxatrone: no CEA studies found
QOL data used from industry submission, they use utility values categorised by age and body mass index based on TTO values. An equation allows estimation of utility values for any BMI. These values are commercial in confidence and not listed in the report. No change in life expectancy is assumed in the baseline analysis. Average age is taken as 40 years and the model time horizon is 20 year (to age 60) Cost per QALY gained ranges from £742 for VBG compared to GB, to £256,856 for GB compared to ASGB
52
22
Obesity - orlistat (N0 22)
March 2001
5 (18)
1 CU from industry submission declared commercial in confidence. No details. 1 CU study reported from lit: QALYs gained 1.601 with orlistat compared to placebo Incremental cost per QALY gained £45,881 compared to placebo (range £19452 - £55,391) No breakdown of QALY detailed. “The authors commented that utilities have been calculated on the basis of published trial results”
31
Obesity - Sibutramine (N0 31)
October 2001
6(6)
Based on data from manufacturer: QOL coefficient 0.00185 per kg lost with sibutramine and 0.00142 with placebo Cost per QALY 10,500
27
Osteoarthritis and rheumatoid arthritis - Cox II inhibitors (N0 27)
July 2001
45
Ovarian cancer (advanced) PLDH (Caelyx) (No.45)
July 2002
Not listed on HTA pages 6 (23)
3
Ovarian cancer - taxanes (N0 3)
May 2000
4 (17)
28
Ovarian cancer - topotecan (N0 28)
July 2001
5 (28)
2 Economic evaluations identified. Both from same trial. Cost minimisation analysis – assumed equivalence. No equivalence of HRQOL is established, however. No utility values/ QALY values in base case analysis Base case results show PLDH dominant with lower mean cost (£2657) and improved mean survival (0.12 life years) Threshold analysis was used to indicate the relative QOL values which would be need to change the conclusions of the CEA. Breast cancer 2 UK studies comparing docetaxel versus paclitaxel found cost per QALY range from £1990 to £2431 1 compared docetaxel to vinorelbine – cost per QALY £14,050 no break down of QALYs Ovarian cancer 1 CUA study found for UK little detail of how QALYs calculated, except that index of HRQOL used cost per QALY for paclitaxel compared with CAP £5433 cost per QALY for paclitaxel compared with carboplatin alone £5273 2 non-UK studies cost per QALY £11,269 and cost per quality-adjusted PFLYG £6860 (high of £10,377) 2 confidential industry submissions 2 similar publications based on same trial no QOL information
53
25
Pancreatic cancer gemcitabine (N0 25)
May 2001
5 (24)
Mean survival gain with gemcitabine: 2.27 months = 0.19 LYG Estimated incremental cost with gemcitabine: £16,543 per LYG (compared with PFI-5FU) Illustration of impact of QOL using Q-TWIST method (quality adjusted time without symptoms or toxicity). Overall survival of patients divided into different health states which are weighted according to QOL in each state: time in clinical benefit; time before progressive disease when not in clinical benefit; time from disease progression to death assigned utilities 1.0, 0.5, and 0.5 – resulting in: QALYs gained 0.148 and incremental cost per QALY £21,088. This is illustrative only, the utility values have no empirical foundation, however, cost per QALY likely to be higher than cost per LYG.
41
Pregnancy - routine anti-D prophylaxis for rhesus negative women (N0 41)
April 2002
Not listed on HTA pages
54
36
Rheumatoid arthritis etanercept and infliximab (No. 36)
March 2002
6(21)
Etanercept (Wyeth) 1 published CEA found: ICER $36,300 1 industry submission: “time-slice spreadsheet model” utilities derived from mapping HAQ scores onto EQ 5D base case £18,948 per QALY (sensitivity analysis range £9942 £48,454) infliximab (Schering Plough) markov model with 6 month cycle Utilities derived from VAS scores. ICER £33,628 per QALY to £36,623 compared to placebo (according to time allowed on infliximab) Sensitivity analysis range £29,008 - £40,766 Birmingham model Assumes no effect on mortality Annual QALY gains estimated as Sulphasalzine: 0.05 Methotrixate: 0.066 Leflunomide: 0.098 Infliximab: 0.120 Etanercept: 0.120 All other DMARDs: 0.05 These are estimated from reduction in HAQ multiplied by a conversion rate of 0.2 QALY gains are 0.172, 0.097 and 0.075 for etan-base, infl-base and etan-infl respectively ICERs are £71,659, £94,798, £41,796 respectively where etanercept and infliximab used as last in sequence. ICERs are £83,000 and £115,000 for infliximab and etanercept if used as 3rd in sequence of DMARDs
43 39
Schizophrenia - atypical antipsychotics (No. 43) Smoking cessation bupropion and nicotine replacement therapy (No. 39)
March 2002
5 (34)
No relevant info reported
March 2002
6 (16)
Model assumes the following 2 life years are saved per quitter (range 1-3) utility values are not reported Based on ratio of 1.35 (QALYs/ life years saved) in one study assume QALYs gained 2.7 (range 1.354.05) Incremental costs per QALY saved are £741-1777 for NRT; £473-1106 for bupropion; £660-1459 for NRT + bupropion
55
49
Ultrasound locating devices for placing central venous catheters (NO 49)
September 2002
4 (35)
5 studies included in DA model, no economic studies, no survival or QOL data in literature cost per restenosis event avoided £1545 extrapolation from 6 month to long term outcome using a conversion based on a study of angioplasty cost per QALY calculated as £6438 baseline QALY gain 0.03 years
1
Wisdom teeth - removal (N0 1)
April 2000
4 (15)
24
Wound care - debriding agents (N0 24)
April 2001
5 (14)
4 decision analyses found 2 DAs incorporated utility values – not reported in much detail -1 patient-derived values: “The results showed that the maximum expected utility of nonextraction (76.96) was better than that for prophylactic third molar surgery (60.25) -1 clinician-estimated values: 69.5 and 63.3 molar retention dominant – lower cost, more effective No QOL/ QALY information
56
REFERENCES Acton JP (1976) Evaluating Public Programs to Save Lives: the Case of Heart Attacks. RAND Corporation, Santa Monica, Report No R950RC. Arrow K, Solow R, Portney PR, Leamer EE, Radner R, Schuman, H (1993) Report of the National Oceanic and Atmospheric Adminstration Panel of Contingent Valuation, Federal Register; 58 (10): 4601-14. Bala M.V., Wood L.L., Zarkin G.A., Norton E.C., Gafni A. and O’Brien B. (1998). Valuing outcomes in health care: a comparison of willingness to pay and quality adjusted life years. Journal of Clinical Epidemiology, 51(8): 667-676. Barrett BJ, Parfrey PS, Foley RN and Detsky AS (1994) An economic analysis of strategies for the use of contrast media for diagnostic cardiac catheterization. Medical Decision Making, 14: 325-335. Beattie J, Covey J, Dolan P, Hopkins, L, Jones-Lee M, Loomes G, Pidgeon N, Robinson A and Spencer A (1998) On the contingent valuation of safety and the safety of contingent valuation: part 1 – caveat investigator. Journal of Risk and Uncertainty 17: 5-25. Berwick DM and Weinstein MC (1985) What do patients value? Willingness to pay for ultrasound in normal pregnancy. Medical Care, 23: 881-93. Bleichrodt H and Quiggin J (1999) Life cycle preferences over consumption and health: when is cost-effectiveness analysis equivalent to cost benefit analysis. Journal of Health Economics, 18: 681-708. Blumenschein K and Johannesson M (1998) The relationship between quality of life instruments, health state utilities and willingness to pay in patients with asthma. Annals of Allergy, Asthma and Immunology, 80(2): 189-194. Blumenschein K, Johannesson M, Yokoyama KK and Freeman PR (2001) Hypothetical versus real willingness to pay in the health sector: results from a field experiment. Journal of Health Economics 20: 441-457. Brooks R (1996) EuroQol: the current state of play. Health Policy 37: 53-72. Bryan S, Roberts T, Heginbotham C and McCallum A (2002) QALY maximisation and public preferences: results from a general population survey. Health Economics, 11: 679-694. Carson RT (1997) Contingent valuation surveys and tests of insensitivity to scope. In Kopp RJ, Pemmerhene W and Schwartz N (eds) Determining the Value of NonMarketed Goods: Economic, Psychological and Policy Relevant Aspects of Contingent Valuation Methods. Kluwer, Boston.
57
Carson RT, Flores NE, Martin KM and Wright JL (1996) Contingent valuation and revealed preference methodologies: comparing the estimates for quasi-public goods. Land Economics 72: 80-99. Carthy T, Chilton S, Covey J, Hopkins, L, Jones-Lee M, Loomes G, Pidgeon N and Spencer A (1999) On the contingent valuation of safety and the safety of contingent valuation: part 2 – the CV/SG ‘chained’ approach. Journal of Risk and Uncertainty 17: 187-213. Chilton SM and Hutchison WG (1999) Do focus groups contribute anything to the contingent valuation process? Journal of Economic Psychology 20: 465-483. Chilton S, Covey J, Hopkins L, Jones-Lee M, Loomes G, Pidgeon N and Spencer A (2002) Public perceptions of risk and preference-based values of safety. Journal of Risk and Uncertainty 25(3): 211-232. Clarke PM (1997) Valuing the benefits of health care in monetrary terms with particular reference to mammographic screening. PhD Thesis, Australian National University, Canberra. Coley CM, Li Y-H, Medsger AR, Marrie TJ, Fine MJ, Kapoor WN, Lave JR, Detsky AS, Weinstein MC and Singer DE (1996) Preferences for home versus hospital care among low-risk patients with community-acquired pneumonia. Archives of Internal Medicine, 156: 1565-1571. Culyer AJ (1971) The nature of the commodity health care and its efficient allocation. Oxford Economic Papers 23: 189-211. Cunningham SJ and Hunt NP (2000) Relationship between utility values and willingness to pay in patients undergoing orthognathic treatment. Community Dental Health, 17: 92-96. Davis RK (1963). Recreation planning as an economic problem. Natural Resources Journal, 3, 239-249. Department for Transport (2002) Highway Economics Note, No.1. Department for Transport, London. Devlin N and Parkin D (2002) Does NICE have a cost-effectiveness threshold and what other factors influence their decisions? A discrete choice analysis. Paper presented to the Health Economists’ Study Group, Brunel University, July 2002. Diener A, O'Brien B and Gafni A (1998) Health care contingent valuation studies: a review and classification of the literature. Health Economics 7: 313-326. Dolan P (2000) The effect of age on health state valuations. Journal of Health Services Research and Policy, 5: 17-21. Dolan P and Cookson R (2000) A qualitative study of the extent to which health gain matters when choosing between groups of patients. Health Policy, 51: 19-30.
58
Dolan P and Edlin R (2001) Is it really possible to build a bridge between cost-benefit analysis and cost-effectiveness analysis? Journal of Health Economics 21: 827-843. Dolan P, Gudex C, Kind P and Williams A (1995) A social tariff for the EuroQol: Results from a UK general population survey. Centre for Health Economics, Discussion Paper 138. Centre for Health Economics, University of York. Donaldson C (1990) Willingness to pay for publicly-provided goods: a possible measure of benefit? Journal of Health Economics 9:103-118. Donaldson C (1998) The (near) equivalence of cost-effectiveness and cost-benefit analysis: fact or fallacy. Pharmaco-Economics 13: 389-396. Donaldson C (1999) Valuing the benefits of publicly-provided health care: does ‘ability to pay’ preclude the use of ‘willingness to pay’? Social Science and Medicine 49: 551-563. Donaldson C and Shackley P (1997) Does “process utility” exist? A case study of willingness to pay for laparoscopic cholecystectomy. Social Science and Medicine 44: 699-707. Donaldson C and Shackley P (2003) Willingness to pay for Health Care. In Scott A, Maynard A and Elliott R (eds) Advances in Health Economics. John Wiley, London. Donaldson C, Birch S and Gafni A (2002) The pervasiveness of the ‘distribution problem’ in economic evaluation in health care. Health Economics 11: 55-70. Donaldson C, Thomas R and Torgerson DJ (1998) Validity of open-ended and payment scale approaches to eliciting willingness to pay. Applied Economics 29:7984. Dubourg WR, Jones-Lee MW and Loomes G (1997) Imprecise preferences and survey design in contingent valuation. Economica 64: 681-702. Evans RG (1984) Strained Mercy: the Economics of Canadian Health Care. Butterworths, Toronto. Gafni A (1991). Willingness to pay as a measure of benefits: relevant questions in the context of public decision making about health care programmes. Medical Care, 29, 1246-52. Gafni A and Feder A (1987) Willingness to pay in an equitable society: the case of the Kibbutz. International Journal of Social Economics 14: 16-21. Garber AM and Phelps CE (1997) Economic foundations of cost-effectiveness analysis. Journal of Health Economics, 16: 1-31. Gegax D, Gerking S and Schulze W (1991) Perceived risk and the marginal valuation of safety. Review of Economics and Statistics 73: 589-596.
59
Gerard K and Mooney G (1993) QALY league tables: handle with care. Health Economics, 2(1), 59-64. Glaser BG and Strauss AL (1967) The Discovery of Grounded Theory. New York: Aldine, Hawthorne. Guria J, Jones W, Jones-Lee MW, Keall M, Leung J and Loomes G (1999) The values of statistical life and prevention of injuries in New Zealand. Report to the New Zealand Transport Safety Authority. Hadorn D (1991) Setting health care priorities in Oregon: cost effectiveness meets the rule of rescue. Journal of the American Medical Association 265: 2218-2225. Hakim Z and Pathak D (1999) Modelling the EuroQol data: a comparison of discrete choice conjoint and conditional preference modelling. Health Economics, 8: 103-116. Hanemann W (1984) Welfare evaluations in contingent valuation experiments with discrete responses: Reply. American Journal of Agricultural Economics 69: 332-341. Hanemann WM (1993) Valuing the environment through contingent valuation. Journal of Economic Perspectives 8: 19-43. Hanley N. Adamowicz V and Wright R (2002) Do choice experiments pass the scope test? A test of scope in a choice experiment examining the benefits of water quality improvements. Paper presented at EAERE/AERE World Congress, Monterey, 2002. Harwood R, Rogers A, Dickinson E, Ebrahim S. Measuring handicap: the London handicap scale, a new outcome measure for chronic disease. Quality in Health Care, 3: 11-16. Henwood KL and Pidgeon NF (1992) Qualitative Research and Psychological Theorizing. British Journal of Psychology, 83: 97-111. Hirth RA, Chernew ME, Miller E, Fendrick MF and Weissert WG (2000) Willingness to Pay for a Quality-Adjusted Life Year: In Search of a Standard, Medical Decision Making, 20: 332-342. Holmes TP and Kramer RA (1995) An independent sample of yea-saying and starting point bias in dichotomous-choice contingent valuation, Journal of Environmental Economics and Management, 29: 121-32. Holsi OR (1969) Content Analysis for the Social Sciences and Humanities. Reading, MA: Addison-Wesley. Hurley, J.F., Holland, M.R., Markandya, A. and Miller, B.G. (2000). Towards Assessing and Costing the Health Impacts of Ambient Particulate Air Pollution in the UK. IOM Research Report, Edinburgh, Institute of Occupational Medicine. Johannesson M (1992) Economic evaluation of lipid lowering – a feasibility test of the contingent valuation approach. Health Policy 20: 309-320.
60
Johannesson M (1995) The relationship between cost effectiveness analysis and cost benefit analysis. Social Science and Medicine 41: 483-489. Johannesson M, Johansson P-O, Kristrom B, Borgquist L and Jonssson B (1993) Willingness to pay for lipid lowering – a health production function approach. Applied Economics 25: 1023-1031. Johannesson M, Jonsson B and Borgquist L (1991) Willingness to pay for antihypertensive therapy – results of a Swedish pilot study. Journal of Health Economics 10: 461-474. Johnson R, Banzhaf M, Desvousges W (2000) Willingness to pay for improved respiratory and cardiovascular health: a multiple-format, stated preference approach. Health Economics, 9: 295-317. Jones-Lee MW (1974) The value of changes in the probability of death or injury. Journal of Political Economy 82: 835-849. Jones-Lee MW (1976) The Value of Life: an Economic Analysis. Martin Robertson, London and University of Chicago Press, Chicago. Jones-Lee MW (1989) The Value of Safety and Physical Risk, Blackwell, Oxford. Jones-Lee MW (1992) Paternalistic altruism and the value of statistical life. Economic Journal. 102: 80-90. Jones-Lee MW, Hammerton M and Philips PR (1985) The value of safety: Results of a national sample survey. Economic Journal. 95: 49-72. Jones-Lee MW, Loomes G and Philips PR (1995) Valuing the prevention of non-fatal road injuries: contingent valuation vs standard gambles. Oxford Economic Papers 47: 676-695. Kennedy C (2002) Revealed preference compared to contingent valuation: radoninduced lung cancer prevention. Health Economics (forthcoming). Klose T (1999) The contingent valuation method in health care. Health Policy 47, 97123. Kobelt G (1997) Economic considerations and outcome measurement in urge incontinence. Urology, 50(6A): 100-107. Krabbe PFM, Essink-Bot M-L and Bonsel GJ (1997) The comparabiltity and reliability of five health-state valuation methods. Social Science and Medicine, 1997, 45: 1641-1652. Kramer RA and Mercer DE (1997) Valuing a global environmental good: U.S. residents’ willingness to pay to protect tropical rain forests, Land Economics, 73: 196210.
61
Krippendorf K (1980) Content Analysis – an Introduction to its Methodology. Beverley Hills, CA: Sage. Kupperman M, Nease RF, Ackerson LM, Black SB, Shinefield HR and Lieu TA (2000) Parents’ preferences for outcomes associated with childhood vaccinations. Pediatric Infectious Disorders Journal, 19: 129-133. Lancaster K (1996) A new approach to consumer theory. Journal of Political Economy, 74: 134-157. Lancaster K (1971) Consumer Demand. Columbia University Press, New York.\ Louviere J, Hensher D, Swait J (2000) Stated preference modelling: Theory, methods and applications. Cambridge University Press, Cambridge, UK.
Loomes G (2002) Valuing life years and QALYs: transferability and convertibility of values across the UK public sector. In Towse A, Pritchard C and Devlin N (eds) Cost Effectiveness Thresholds: Economic and Ethical Issues. King’s Fund and Office of Health Economics, London. Lundberg L, Johannesson M, Silverdahl M, Hermansson C and Lindberg M (1999) Quality of life, health state utilities and willlingness to pay in patients with psoriasis and atopic eczema. British Journal of Dermatology, 141: 1067-1075. McFadden D (1973) Conditional logit analysis of qualitative choice behaviour. University of California at Berkeley, California. Mishan EJ (1971) Evaluation of life and limb: a theoretical approach. Journal of Political Economy 79: 687-705. Mooney (1977) The Valuation of Human Life. Macmillan, London. Mooney (1994) Key Issues in Health Economics. Harvester Wheatsheaf, London. Netten A, Ryan M, Skatun D and Smith P (2002) The development of a measure of social care outcome for older people. PSSRU Discussion Paper 1690, University of Kent at Canterbury. Nord E, Pinto JL, Richardson J, Menzel P and Ubel P (1999) Incorporating societal concerns for fairness in numerical valuations of health programmes. Health Economics 8: 25-39. O’Brien B and Gafni A (1996) When do the “dollars” make sense? Toward a conceptual framework for contingent valuation studies in health care, Medical Decision Making, 16, 288-299. O'Brien B and Viramontes J (1994) Willingness to pay: a valid and reliable measure of health state preference? Medical Decision Making, 14: 289-297.
62
Olsen JA and Donaldson C (1998) Helicopters, hearts and hips: using willingness to pay to set priorities for public sector health care programmes. Social Science and Medicine, 46: 1-12. Olsen JA and RD Smith (2001) Theory versus practice: a review of ‘willingness-topay’ in health and health care. Health Economics 10, 39-52. Pauly M (1995) Valuing health care benefits in money terms. In F Sloan (ed) Valuing Health Care: Costs, Benefits and Effectiveness of Pharmaceuticals and Other Medical Technologies. Cambridge University Press, Cambridge. Phelps C and Mushlin A (1991) The (near) equivalence of cost effectiveness and cost benefit analysis. International Journal of Technology Assessment in Health Care, 7: 12-21. Protière C, Donaldson C, Luchini S, Moatti JP and Shackley P (2003) The impact of information on non-health attributes on willingness to pay for multiple health care programmes. Social Science and Medicine, submitted. Ready RC, Buzby JC and Hu D (1998) Differences between continuous and discrete contingent value estimates, Land Economics, 72: 397-411. Ryan M (1999) Measuring benefits in health care: the role of conjoint analysis. Paper presented at International Health Economics Association, Rotterdam, June 1999. Ryan M and Farrar S (2000) Using conjoint analysis to elicit preferences for health care. BMJ 320: 1530-1533. Ryan M and Gerard K (2003) Using discrete choice experiments to value health care: current practice and future prospects. Applied Health Economics and Policy Analysis (forthcoming). Ryan M and Hughes J (1997) Using conjoint analysis to assess women’s preferences for miscarriage management. Health Economics 6: 261-273. Schelling T (1968) The life you save may be your own. In Chase SB jr (ed) Problems in Public Expenditure Analysis. Washington DC, Brookings Institution. Schkade DA and Payne JW (1994) How people respond to contingent valuation questions: a verbal protocol of willingness to pay for environmental protection. Journal of Environmental Economics and Management 26: 88-109. Shiell A, Seymour K, Hawe P, Cameron S (2000) Are preferences over health states complete? Health Economics, 9(1):47-55. Smith R (2001) The relative sensitivity of willingness to pay and time trade off to changes in health status: an empirical investigation. Health Economics, 10: 487-497.
63
Smith VK (1983) The role of site and job characteristics in hedonic wage models. Journal of Urban Economics 13: 296-321. Sorum PC (1999) Measuring patient preferences by willingness to pay to avoid: the case of acute otitis media. Medical Decision Making, 19: 27-37. Swan JS, Fryback DG, Lawrence WF, Katz DA, Heisey DM, Hagenauer ME, Selzer PM and Jacobson BK (1997). MR and conventional angiography: work in progress towards assessing utility in radiology. Academic Radiology, 4: 475-482. The Urban Institute (1991) The costs of highway crashes. Report number FHWA-RD91-055, The Urban Institute, Washington DC. Final Report. Washington DC. Thompson M (1986) Willingness to pay and accept to cure chronic disease. American Journal of Public Health, 76: 392-396. Torrance GW (1986) Measurement of health state utilities for economic appraisal. Journal of Health Economics, 5: 1-30. Tsuchiya A (1999) Age-related preferences and age weighting health benefits. Social Science and Medicine, 48(2): 267-276. Tsuchiya A (2000) QALYs and ageism: Philosophical theories and age weighting. Health Economics, 9(1): 57-68. Viscusi WK and Moore MJ (1989) Rates of time preference and valuation of the duration of life. Journal of Public Economics 38: 297-317. Voruganti LNP, Awad AG, Oyewumi LK, Cortese L, Zirul S and Dhawan R (2000) Assessing health utilities in schizophrenia: a feasibility study. Pharmacoeconomics, 17: 273-286. Williams A (1985) Economics of coronary artery bypass grafting. British Medical Journal, 291: 326-329. Williams A (1997) Intergenerational equity: an exploration of the ‘fair innings’ argument. Health Economics 6: 117-132. Zethraeus N (1998) Willingness to pay for hormone replacement therapy. Health Economics, 7: 31-38. Zethraeus N, Johannesson M, Henriksson P and Strand RT (1997) The impact of hormone replacement therapy on quality of life and willingness to pay. British Journal of Obstetrics and Gynaecology, 104: 1191-1195. Zwerina K, Huber J, Kuhfield W (1996) A general method for constructing efficient choice designs. Working paper, Fuqua School of Business, Duke University.
64
