EXPLORING STATISTICAL UNCERTAINTY OF HEALTH INTERVENTION EFFECTS IN DISCRETE CHOICE EXPERIMENTS: A FLU VACCINE CASE STUDY
Author(s)
Carina Oedingen, PhD1, Stella Maria Marceta, MSc1, Matthew Robson, PhD1, Vahid Moghani, PhD1, Tom Van Ourti, PhD1, Joffre D. Swait, PhD1, Esther de Bekker-Grob, PhD1, Raf Van Gestel, PhD2, Jorien Veldwijk, PhD1.
1Erasmus University Rotterdam, Rotterdam, Netherlands, 2KU Leuven, Leuven, Belgium.
1Erasmus University Rotterdam, Rotterdam, Netherlands, 2KU Leuven, Leuven, Belgium.
OBJECTIVES: Inputs for attributes and attribute levels in preference studies, such as estimates of risks and effectiveness, are subject to various kinds of statistical uncertainty. To date, such uncertainty is rarely incorporated, and it remains unclear whether and how it can be operationalized. This study investigates the effect of incorporating statistical uncertainty into a discrete choice experiment (DCE) on comprehension and study outcomes.
METHODS: A DCE measuring preferences for flu vaccination among elderly (aged 60 or older) individuals in the Netherlands (n=526) served as a case study. The DCE included five attributes: vaccine effectiveness, risk of mild side effects, risk of severe side effects, protection duration, and incubation time. Statistical uncertainty was incorporated for the attribute ‘vaccine effectiveness’ only. Mean vaccine effectiveness levels of 20%, 30%, 60% and 70% were described numerically and graphically, with statistical uncertainty expressed as ranges of +/-5%, +/-10% and +/-15%. Multinomial logit and latent class models were used for the preference analyses, and comprehension, perceived understanding, and use of heuristic decision-making were investigated.
RESULTS: Respondents generally understood the mean vaccine effectiveness, but many had difficulty interpreting or applying the exact uncertainty ranges. Overall, higher mean effectiveness was associated with higher utility, whereas disutility was associated with higher statistical uncertainty. Incorporating statistical uncertainty also increased respondents’ attention to the vaccine effectiveness attribute, with fewer respondents ignoring this attribute, but also led to increased dominant decision-making.
CONCLUSIONS: Incorporating statistical uncertainty into DCEs is feasible and could be valuable in situations in which attributes are considered as uncertain. Future DCEs, especially those used to inform market forecasts or policy decisions as well as clinical context, should consider including statistical uncertainty in the design of their attribute levels to closely mimics real-world decision-making.
METHODS: A DCE measuring preferences for flu vaccination among elderly (aged 60 or older) individuals in the Netherlands (n=526) served as a case study. The DCE included five attributes: vaccine effectiveness, risk of mild side effects, risk of severe side effects, protection duration, and incubation time. Statistical uncertainty was incorporated for the attribute ‘vaccine effectiveness’ only. Mean vaccine effectiveness levels of 20%, 30%, 60% and 70% were described numerically and graphically, with statistical uncertainty expressed as ranges of +/-5%, +/-10% and +/-15%. Multinomial logit and latent class models were used for the preference analyses, and comprehension, perceived understanding, and use of heuristic decision-making were investigated.
RESULTS: Respondents generally understood the mean vaccine effectiveness, but many had difficulty interpreting or applying the exact uncertainty ranges. Overall, higher mean effectiveness was associated with higher utility, whereas disutility was associated with higher statistical uncertainty. Incorporating statistical uncertainty also increased respondents’ attention to the vaccine effectiveness attribute, with fewer respondents ignoring this attribute, but also led to increased dominant decision-making.
CONCLUSIONS: Incorporating statistical uncertainty into DCEs is feasible and could be valuable in situations in which attributes are considered as uncertain. Future DCEs, especially those used to inform market forecasts or policy decisions as well as clinical context, should consider including statistical uncertainty in the design of their attribute levels to closely mimics real-world decision-making.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
P46
Topic
Methodological & Statistical Research, Patient-Centered Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Vaccines