Issues in the Design of Discrete Choice Experiments

The use of preference-elicitation tasks—in particular, discrete choice experiments (DCEs)—in health economics has grown significantly in recent decades [1]. The most widely used DCE approach asks respondents to consider a series of hypothetical choices between alternatives (here called choice tasks), and to specify which alternative they prefer. The use of choice tasks in other areas—especially psychology, transportation, marketing and agriculture—has a more established history. Health preference studies have been conducted for about as long [2, 3], but not to the same extent; the relatively late uptake in preference evidence in health is surprising in some regards as patient and population values concerning health have always been key components of a range of questions from health policy to clinical practice, and often cannot be directly observed, a problem exacerbated by the lack of a perfectly competitive market [4]. Though there is broad consensus on the value patients or the population place on health matters in decision-making, the methods for including them in a way that is reliable are debated. A significant issue with the conduct of such experiments is how best to construct the choice tasks to produce policyrelevant and reliable value estimates. If, for simplicity, the task has two alternatives (i.e., paired comparison [5]), which alternatives should be presented head-to-head? The risk of picking the wrong combinations are that values for some alternatives either cannot be estimated at all, or that they are estimated with an unacceptably low level of precision. This topic is of course not unique to health, and we should be cognisant of the work being conducted in other fields using similar methods. Conversely, we also believe that the design of health preference studies requires specific consideration to reflect the nature of the questions, and the provision of results, that best informs decision makers. This paper provides a summary of a panel discussion from a DCE design symposium at the International Academy of Health Preference Research (IAHPR) 2018 meeting in Hobart, Australia on September 28, 2018. This paper is one of a series of manuscripts reflecting on key issues discussed at IAHPR meetings [6, 7]. At the start of the symposium, each panellist presented lessons learned from their own experience: John Rose (JR) A unified theory of experimental design for stated choice studies Deborah J. Street (DJS) What can simulations tell us about DCE design performance? Marcel F. Jonker (MFJ) Individually adaptive D-efficient DCE designs Paul Hansen (PH) The PAPRIKA method: A full factorial DCE involving pairwise rankings of all possible attribute combinations Benjamin M. Craig (BMC) Experience-based methods for DCE designs This article is part of the topical collection on ‘‘From the International Academy of Health Preference Research’’.