The use of risk-based discrete choice experiments to capture preferences over health states

Discrete choice experiments (DCEs) allow a number of characteristics to be traded-off against one another. An overriding methodological challenge faced is how best to apply DCEs to questions involving those attributes commonly used in value elicitation exercises such as risk, time (Bansback et al. 2012) and numbers treated (Robinson et al, 2010). Flynn (2010) concluded that in developing the methods, it was important to understand more fully the preferences of individual respondents. The study reported here sets out to provide such insights by enhancing a DCE design with additional questions that allow utility values to be derived at the individual level also. The DCE presented respondents with eight pairwise risky choices to estimate aggregate utility values for three EQ-5D health states, ranging from mild to severe. The design allowed the elicitation of utility values for worse-than-dead states. Risk was represented using the stimulus used by EuroVaQ (http://research.ncl.ac.uk/eurovaq/). Three main devices were used to collect additional individual level data. Firstly we included six additional DCE questions that were not used to estimate the aggregate DCE model but allowed the utility value of one health state to be estimated at the level of the individual. These six questions provided more extensive tests of the internal consistency of the pairwise choices undertaken in the DCE. Secondly, respondents were asked three questions where the risk in one of the two treatments was fixed, and they set the risk of the other treatment (a modified SG question). These questions then allowed us to estimate utility values for all three health states. Finally, we collected respondents risk attitudes using Kuilen and Wakker's 2011 measure. We collected data on a convenient sample of 59 students studying Economics or Geography at the University of London and Exeter in 2011/12. Preliminary results show that 22 of the 59 respondents gave a series of DCE responses that were internally inconsistent. We report here the implications of the results for the inclusion of risk as an attribute in DCEs and for preference elicitation more broadly.