Does Changing the Way a Discrete Choice Experiment (DCE) is Presented to Respondents Affect Results? an Investigation in the Context of Health Using Between-Subject Designs

Abstract

Discrete choice experiments (DCEs) are a popular stated preference technique used in health. A major challenge in health related DCEs is that they often involve terminology and concepts that may be unfamiliar to respondent. Therefore, it is important for DCEs to include explanations of the concepts and terminology used where needed. This raises questions about the different ways information in a DCE is presented to respondents. Most importantly, does the type of information and level of information provided in a DCE impact results?

In order to answer this question, this thesis included a scoping review and three empirical studies that add to the literature. The scoping review summarised DCE studies that investigated the impact of either the amount of information or the presentation style of the DCE or both on the choices of respondents. Studies were included if they allocated respondents to two or more arms to test the impact of differences in the information provided to respondents, either explicitly (e.g., providing more detailed information in one arm than in the other) or implicitly (e.g., by having more options in each choice set). The review showed that the impact of presentation differences on DCE findings was not consistent across studies and this begs the question of whether it is desirable or possible to design DCEs with conclusions that are robust to how the DCE is presented to respondents.

The first empirical study investigated patient preferences for features of an assessment tool for peripheral neuropathy, a possible side effect of cancer treatment. The second study investigated the same topic but this time in a general population sample. The aim was to investigate the differences in preferences between the two samples. Another aim was to understand whether increasing the level of information provided to the general population sample had an impact on results. This was investigated by splitting the general population sample into two arms, where one arm received more information compared to the other. The presentation format was also different between arms, with one arm receiving static images and the other a short video as well as GIFs i.e. moving images on loop. The study found that providing extra information and the use of different presentation formats did not lead to significant preference differences between the two general population arms. The patient and general population were found to have similar preferences. Although differences were noted in terms of scale. The scale difference was found to be somewhat smaller between the general population arm that received more information and the patient sample compared to that between the general population arm that received only basic information and the patient sample.

The third empirical study investigated different anchoring methods and its impact on the valuation of the EQ-5D-5L, a widely used quality of life measurement instrument. Respondents were shown choice sets that included three options; two health states and a third option which was immediate death or full health i.e. the anchoring option. The study found that health state utilities were sensitive to the wording used in the third option of the choice sets; in particular, choice sets with immediate death as the third option consistently produced a narrower range of utilities than did those with full healthas the third option.

Evidence from this thesis demonstrates that information in a DCE may impact the resulting preferences, but that the extent of this impact varies by context.