Abstract
INTRODUCTION: To assess convergent validity of stated preference methods in studies where they were used to elicit patient preferences for informing medical product decisions. METHODS: In four studies, two stated preference methods were used to elicit preferences of patients with neuromuscular diseases (NMD; n = 140, Discrete Choice Experiment [DCE] and Best-Worst Scaling [BWS] case 2), diabetes (n = 495, DCE and swing weighting [SW]), myocardial infarction (MI; n = 335, DCE and BWS case 1), and rheumatoid arthritis (RA; n = 982, DCE and probabilistic threshold technique [PTT]). In each study, results of the two methods were compared using a normalized preference measure for which confidence intervals (CIs) were estimated using nonparametric bootstrapping of 500 samples. Normalized preference measures comprised of mean relative attribute importance weights (NMD and diabetes studies), attribute uptake probability (MI study), or maximum acceptable risk (RA study). RESULTS: In all four studies, attribute ranking showed similar patterns between DCE and other methods for the most important attributes. The same attribute had highest importance in three out of four studies. Significant differences were found in ranges of normalized preference measures of each study between DCE and the other methods: 4.1–43.4 versus 8.9–24.7 for DCE and BWS case 2 in NMD; 3.8–49.7 versus 11.9–16.8 for DCE and SW in diabetes; 2.0–85.5 versus 0.2–69.0 for DCE and BWS case 1 in MI; -3.5–49.2 versus 1.1–18.1 for DCE and PTT in RA. CONCLUSIONS: Preferences differed significantly between DCE and other preference methods implying limited convergent validity. The substantially larger ranges in normalized outcome measures in DCE compared to other methods, are likely due to differences in mechanics and bias related to the methods. Since none of the methods is considered the golden standard for measuring stated preferences as true preferences are unknown, further studies are necessary to compare stated preference methods, determine internal validity and data quality, and potentially measure external validity.