Abstract
OBJECTIVE: We aimed to empirically compare maximum acceptable risk results estimated using both a discrete choice experiment (DCE) and a probabilistic threshold technique (PTT). METHODS: Members of the UK general public (n = 982) completed an online survey including a DCE and a PTT (in random order) measuring their preferences for preventative treatment for rheumatoid arthritis. For the DCE, a Bayesian D-efficient design consisting of four blocks of 15 choice tasks was constructed including six attributes with varying levels. The PTT used identical risk and benefit attributes. For the DCE, a panel mixed-logit model was conducted, both mean and individual estimates were used to calculate maximum acceptable risk. For the PTT, interval regression was used to calculate maximum acceptable risk. Perceived complexity of the choice tasks and preference heterogeneity were investigated for both methods. RESULTS: Maximum acceptable risk confidence intervals of both methods overlapped for serious infection and serious side effects but not for mild side effects (maximum acceptable risk was 32.7 percent-points lower in the PTT). Although, both DCE and PTT tasks overall were considered easy or very easy to understand and answer, significantly more respondents rated the DCE choice tasks as easier to understand compared with those who rated the PTT as easier (7-percentage point difference; p < 0.05). CONCLUSIONS: Maximum acceptable risk estimate confidence intervals based on a DCE and a PTT overlapped for two out of the three included risk attributes. More respondents rated the DCE as easier to understand. This may suggest that the DCE is better suited in studies estimating maximum acceptable risk for multiple risk attributes of differing severity, while the PTT may be better suited when measuring heterogeneity in maximum acceptable risk estimates or when investigating one or more serious adverse events.