Using a discrete choice experiment to estimate health state utility values
Introduction
A key input to the quality adjusted life year (QALY) is the relative value of time spent in different health states (Torrance, 1986). Currently, the dominant methods for eliciting these values are the standard gamble (SG) and the time trade off (TTO), cardinal preference techniques which produce values anchored by full health and being dead (Brazier et al., 2007). The TTO was originally designed as a simpler alternative to the SG (Torrance et al., 1972). Its premise is that the utility change associated with a decrement in health status is determined by valuing the amount of life expectancy an individual is prepared to sacrifice to avoid experiencing the decrement that leaves overall utility unchanged. Accordingly, for a suboptimal health state preferred to being dead, respondents are asked a series of questions regarding whether they would prefer to live in the health state for t years then die, or live in full health for h years (where h < t) then die. For a suboptimal health state considered ‘worse than dead’ (WTD), a series of questions regarding whether respondents would prefer to live t years in the health state followed by h years in full health or die immediately. In both cases, h is varied through a predetermined iterative process until the respondent is indifferent between the two options. In the most common TTO procedure, tradeoffs are illustrated to respondents using a visual board, as exemplified in Fig. 1.
Although the TTO is generally regarded as being simpler for respondents to complete than the SG, there is still a concern that the tasks involved are still too cognitively demanding for certain populations, resulting in response inconsistencies and subsequent data exclusions, which limit the representativeness of the values obtained (Bosch et al., 1998, Craig et al., 2009a). Since the TTO is widely used to obtain valuations for generic health state descriptions through surveys of representative samples of the general public, any data exclusions are of particular concern. Concerns have also been raised about using different tasks for states better and WTD (Lamers, 2007), and the way TTO data has conventionally been modelled (Craig and Busschbach, 2011). An alternative elicitation method that some authors have argued may be simpler than the conventional iterative TTO task is the discrete choice experiment (DCE) (Louviere and Woodworth, 1983), a technique from market research that over the last 10 years has been applied to valuing preferences for health states (e.g. Hakim and Pathak, 1999). Conventional DCE tasks present two or more profiles, each made up of levels of attributes selected from a descriptive system (Louviere et al., 2000). Unlike the conventional TTO, DCEs require respondents to simply indicate that option A is preferred to B, without going through an iterative process of identifying the point at which the respondent is indifferent between A and B. DCE tasks are generally considered simple to complete, and they are often conducted without an interviewer through postal or on-line surveys, but this is dependent on characteristics of the specific task including the number of attributes.
Notably, since the TTO also asks participants to make a series of pairwise choices between two discrete options, the TTO could be considered as a form of DCE. In fact, if an attribute were to be included in a DCE to represent years of survival in a health state, the choice task would closely resemble the TTO, in that it would ask respondents to choose between health state profiles containing a health state description and the life years that would be spent in that health state (Fig. 1). A primary motivation to use such a DCE design would be to produce values amenable to QALY calculations, as has been previously proposed (Ryan et al., 2006, Viney et al., 2007, Coast et al., 2008, Flynn et al., 2008).
Through the conditional logit model, and its variations, DCE data can provide information on the relative preference of one health state over another. However, the scale is not anchored on the health utility scale where 0 is the value of being dead and 1 is the value of full health, meaning DCE data cannot be directly incorporated into QALY calculations. While previous studies have successfully anchored DCE results on the health utility scale, their methods have depended on problematic assumptions (Bosch et al., 1998, Ryan et al., 2006, Burr et al., 2007, Ratcliffe et al., 2009, Flynn et al., 2008). We devise a DCE that consists of choices between health profiles, i.e. a health state with a specified duration, herein referred to as DCETTO, which may be able to overcome theoretical challenges and avoid making questionable assumptions, but has not yet been attempted. We aimed to determine whether such a design is advantageous over the conventional TTO and to establish a method for anchoring its results on the health utility scale.
The purpose of this study was to evaluate the novel DCETTO, to anchor the results on the health utility scale using appropriate methods, and to compare the process with the conventional iterative TTO. The survey methods are described below in Section 2, while Section 3 details the econometric modeling and anchoring assumptions utilised in the study. The results of the DCETTO and TTO exercises are described in Section 4. Finally, the implications of these results for future elicitation of health state values are discussed in Section 5.
Section snippets
Survey and elicitation tasks
A web survey was conducted asking respondents to complete a series of TTO and DCETTO tasks. Health states in the survey were described using the EQ-5D descriptive system (Brooks, 1996), which consists of five attributes (mobility, self-care, usual activities, pain/discomfort and anxiety/depression), with three possible levels for each attribute. Level 1 refers to the best level in each attribute – so health state 11111 refers to full health and 33333 refers to the worst health state possible in
Time trade off
TTO responses were analysed using the episodic random utility model (ERUM) (Craig and Busschbach, 2009, Craig and Busschbach, 2011), where the value of a health state depends on its duration, tij:where i = 1, 2, …, n represents individuals and j = 1, 2, …, m represents the different health states shown to each respondent. The dependent variable, Vij is the value for health state j valued by each respondent i. xij is a
The sample
A sample of 4189 members of the market research panel was initially invited by email to participate in the survey. Of these, 1400 (33%) consented to begin the survey and 1157 (83% of those who consented) completed both the TTO and DCETTO. Of the 689 respondents who were randomised to begin with the TTO exercise, 95 (14%) completed one or no tasks. Of the 711 who began with the DCETTO exercise, 43 (6%) failed to complete all eight DCETTO tasks. Among the 1355 respondents who began the TTO
Discussion
The novel contribution of this paper is the development and interpretation of a DCE task that includes a life years attribute, specifically for the purposes of obtaining values anchored on the health utility scale for use in QALY calculations. The values from the method appear robust, with estimated coefficients that are statistically significant, logically consistent, and with expected signs. The primary motivation of the study was to evaluate the novel DCETTO by anchoring the results on the
References (41)
- et al.
Approximation of Bayesian efficiency in experimental choice designs
Journal of Choice Modelling
(2008) EuroQol: the current state of play
Health Policy
(1996)- et al.
Valuing the ICECAP capability index for older people
Social Science & Medicine
(2008) - et al.
Keep it simple: ranking health states yields values similar to cardinal measurement approaches
Journal of Clinical Epidemiology
(2009) - et al.
Using discrete choice experiments to estimate a preference-based measure of outcome—an application to social care for older people
Journal of Health Economics
(2006) Eliciting GPs’ preferences for pecuniary and non-pecuniary job characteristics
Journal of Health Economics
(2001)- et al.
Discrete choice modeling for the quantification of health states: the case of the EQ-5D
Value in Health
(2010) Measurement of health state utilities for economic appraisal
Journal of Health Economics
(1986)- et al.
Estimating general-population utilities using one binary-gamble question per respondent
Medical Decision Making
(1998) - et al.
Measuring and Valuing Health Benefits for Economic Evaluation
(2007)
Developing a preference-based Glaucoma Utility Index using a discrete choice experiment
Optometry and Vision Science
The episodic random utility model unifies time trade-off and discrete choice approaches in health state valuation
Population Health Metrics
Modeling ranking, time trade-off, and visual analog scale values for EQ-5D health states: a review and comparison of methods
Medical Care
Toward a more universal approach in health valuation
Health Economics
Logical inconsistencies in survey respondents’ health state valuations – a methodological challenge for estimating social tariffs
Health Economics
The generalized multinomial logit model: accounting for scale and coefficient heterogeneity
Marketing Science
Rescaling quality of life values from discrete choice experiments for use as QALYs: a cautionary tale
Population Health Metrics
Modelling the EuroQol data: a comparison of discrete choice conjoint and conditional preference modelling
Health Economics
Cited by (187)
Examining interrater agreement between self-report and proxy-report responses for the quality of life-aged care consumers (QOL-ACC) instrument
2024, Journal of Patient-Reported OutcomesComparison of four value sets derived using different TTO and DCE approaches: application to the new region-specific PBM, AP-7D
2024, Health and Quality of Life Outcomes