Scholars@Duke publication: Means Can Be Deceiving: Comparing and Contrasting Risk Tolerance Estimates From a Discrete-Choice Experiment and a Threshold Technique Exercise.

Means Can Be Deceiving: Comparing and Contrasting Risk Tolerance Estimates From a Discrete-Choice Experiment and a Threshold Technique Exercise.

Publication , Journal Article

Sutphin, J; Wallace, MJ; Reed, SD; Duke PrefER and FDA Patient Preference Methods Working Group

Published in: Value Health

September 25, 2025

OBJECTIVES: This study compared quantitative measures of risk tolerance between 2 preference-elicitation methods: a discrete-choice experiment (DCE) and a probabilistic threshold technique (TT) exercise. METHODS: A survey offered benefit-risk trade-offs pertaining to devices used in revascularization procedures for peripheral artery disease. Survey design features included alternating the sequence of DCE and TT exercises, testing 2 risk-communication approaches, and using 2 DCE experimental designs. The risk tolerance metric was the maximum-acceptable risk (MAR) increase in 5-year mortality, above an 8% baseline risk, that patients would accept to choose a device offering lower repeat-procedure risks. DCE data were analyzed with mixed-logit models, and TT data were analyzed with interval regressions. Sensitivity analyses were conducted to examine the impacts of survey design features. RESULTS: For the full sample (N = 249), MARs from the DCE and TT differed by <1 percentage-point (DCE: 13.0%; TT: 13.8%). MAR estimates were not significantly influenced by the risk-communication approach or experimental design; however, order of DCE and TT mattered. After removing the influence of DCE and TT order, the mean MARs still differed by <1 percentage point (MAR among those completing DCE first [n = 123]: 14.3%; MAR among those completing TT first [n = 126]: 13.7%). Although convergent validity was found at the sample level, discordance (>2 percentage points) between individual-level DCE MAR and TT MAR was observed for approximately half (48.2%) of the sample. CONCLUSIONS: Although there was concordance between DCE and TT mean MAR estimates at the sample level, at the individual level, only half of the respondents had an absolute difference of 2 percentage-points or less.