Quantitative stated preferences elicitations with discrete choice experiments (DCEs) is increasingly used to inform decision making, such as benefit-risk assessments by regulators, reimbursement decisions by health technology assessment agencies, and to guide patient-level clinical decisions. A common yet important challenge in DCE design is determining the appropriate sample size, given constraints such as scope, budget, and timelines.
This paper lays out the key issues in defining sample size in DCEs and critically examines existing calculations and ‘rules of thumb’ for estimating sample size. The paper also outlines specific considerations for DCEs conducted in health and/or with pharma where the key objective is to estimate the effects of changes in attributes on utility. As utility functions are scale invariant, interpretation requires normalization using ratios, and the exact effect size depends on the final utility function being estimated. In addition, the magnitude of utility is confounded with the error variance. Error variance and effects sizes also depend on the population, context, and study design properties. As such, a priori effect sizes are generally unknown without a highly-comparable prior study. In addition to these technical aspects, the sample in preference studies is often constrained by challenges in recruiting a large number of patients because of practical difficulties in recruitment. The paper concludes by suggesting a focus on one particular value or comparison to calculate the sample size required. Other methods (such as threshold technique) could be used with more success to formulate sample size calculation in preference studies.
