Discrete choice experiments (DCEs) are robust stated-preference methods frequently used to estimate maximum acceptable risk (MAR) as a secondary outcome. However, DCEs provide sample-level estimates and explaining preference heterogeneity for MARs based on participant characteristics can be difficult. The study objective was to compare the capability of a DCE and a probabilistic threshold technique (PTT) to identify preference heterogeneity among MARs for preventive rheumatoid arthritis (RA) treatment.
Participants from 3 countries (United Kingdom (UK), Germany, and Romania, n = 2959) completed a DCE and PTT in random order. Participants made choices between treatments that reduced chance of developing RA but increased chance of three risks (mild and serious side effects, serious infection). For the PTT, interval regressions estimated MARs that accounted for age, education, numeracy, literacy, and RA family history. For the DCE, random parameters logit (RPL) models were used to calculate MARs for subgroups in which heterogeneity was identified in the PTT.
The PTT identified preference heterogeneity for numeracy, literacy, and family history. Regarding these characteristics, the PTT identified statistically significantly different MARs (p<0.05) for at least one risk in at least two countries. The DCE identified preference heterogeneity for the chance of serious infection between UK participants with low vs. high numeracy (p<0.05). Using the DCE, no statistically different MARs were identified for other combinations of participant characteristics, risks, or countries.
The PTT identified preference heterogeneity in MARs for more participant characteristics by directly incorporating participant characteristics in the regression model. When attempting to estimate MARs, PTT may partially overcome challenges with stratified DCE models, particularly if analyses such as latent class analysis are not feasible or desirable. Further research is needed to confirm the findings in this case studies and to explore which method most accurately identify true underlying preference heterogeneity are needed.
© 2021 Published by Elsevier Inc.