Abstract

Objectives

The study objectives were (1) to demonstrate the feasibility of constructing a stated-preference evidence base and its use to quantify patients’ consensus risk tolerance for treatment efficacy and (2) to use the evidence base to inform a new, parsimonious choice experiment to test a hypothesis for which there is no evidence-base information.

Methods

Nine original data sets from 5 discrete-choice-experiment studies that included inflammatory bowel disease symptom-remission and serious-infection risk attributes were obtained, totaling 2247 respondents and 25 017 choice questions. All 9 data sets were pooled and fused in a single scale-adjusted, random-parameters logit, latent-class model describing risk-tolerant and risk-averse class preferences plus a statistically uninformative class. We used a 7-data set fusion model to predict maximum acceptable risk for 2 holdout data sets.

Results

Class-membership probabilities for the risk-tolerant, risk-averse, and statistically uninformative classes were 0.53, 0.35, and 0.12, respectively. Consensus maximum acceptable 1-year risks of serious infection for 1 month of symptom remission were 9.5% (8.5, 10.6) and 5.8% (4.5, 7.1) for the risk-tolerant and risk-averse preference classes, respectively. The 7-data set fusion model performed well for combined inflammatory bowel disease out-of-sample predictions but predicted disease-specific values less accurately.

Conclusions

Maturation of the stated-preference literature offers opportunities to treat multiple quantitative preference studies similar to how multiple clinical studies are evaluated to estimate consensus effect sizes. There is significant value in developing and utilizing stated-preference evidence bases to provide benefit-transfer values, as well as to identify information gaps and inform efficient de novo study designs to close those gaps.