Purpose: To pilot test discrete choice experiment (DCE) and constant sum paired comparison (CSPC) online questionnaires for eliciting societal preferences for the allocation of healthcare resources.
Method: Respondents were asked to allocate a fixed budget between two patient groups, described by attributes including age, health and life expectancy with/without treatment, the number of patients and potential quality-adjusted life years (QALYs) gained. The DCE asked respondents to allocate the entire budget to their preferred group, while the CSPC asked respondents to allocate budget percentages between the groups. Each questionnaire presented ten choice tasks, including one repeated task. Because completion rates were expected to be lower with CSPC, 60% of individuals were randomly assigned the CSPC and 40% the DCE. Respondents were also asked to rate the importance of each attribute in their choices, and the difficulty of understanding and of answering the tasks. Response behaviours in the questionnaires were compared in terms of the consistency of responses in a repeated task, and the correlation between tasks and variance in responses to identify learning (decreasing variance) or fatigue (increasing variance) effects.
Result: A significantly greater proportion of individuals completed the DCE (154/256=60%) compared with the CSPC (150/348=43%), p<0.001. Among individuals completing a questionnaire, there was no significant difference in the proportion that rated the tasks "somewhat difficult" or "extremely difficult" to understand (DCE=12%, CSPC=13%; p=1) or to answer (DCE=65%, CSPC=66%; p=.94). Responses to the repeated DCE task were not significantly different (p=.22), but individual differences in the repeated CSPC budget allocations were significant (mean difference=7.6%; p<0.001). There were no significant correlations between task and variance for either the DCE or CSPC. Fewer than 10% of all CSPC allocations explicitly equalised QALYs, patients or budgets. Eighteen percent of allocations maximised resources to one particular group and 11% of CSPC respondents maximised in >50% of tasks.
Conclusion: Despite similarities in the ability of respondents completing a questionnaire to understand and answer the DCE and CSPC tasks, the significantly lower completion rate suggests that CSPC was less acceptable in some aspect(s). Consistency across tasks was also better with DCE. However, CSPC can reveal distributional preferences that DCE can not. Better response rates and consistency (‘respondent efficiency’) with DCE must be weighed against potentially richer preference data with CSPC.