Small differences in EQ-5D-5L health utility scores were interpreted differently between and within respondents

OBJECTIVES: This study aims to determine how population-based health-utility score (HUS) differences reflect individuals' health preferences using responses from the Canadian EQ-5D-5L Valuation Study, including time trade-off (TTO) and discrete-choice experiment (DCE) tasks (n = 1073). STUDY DESIGN AND SETTING: Cardinal TTO responses were transformed into pairwise comparisons to yield ordinal TTO responses. We investigated how EQ-5D-5L HUS differences differ from participants' stated cardinal preferences, and determined the smallest HUS difference that may be expected to represent participants' ordinal preferences. RESULTS: HUS differences near zero have 30.6% (95% confidence interval: 29.1-31.9%) probability of representing a tie in individuals' TTO values. Differences in EQ-5D-5L HUS of -0.054 (-0.071 to -0.029) and 0.047 (0.026-0.076) maximized the sensitivity and specificity of discriminating transitions to worse/better health states. For small HUS differences of ±0.03 to ±0.07, the magnitude of respondents' average TTO difference on the cardinal scale was 0.17 and 0.35 whether ties were included or excluded, respectively. Absolute HUS differences between 0.042 and 0.062 had a 50% probability of representing respondents' ordinal preferences. CONCLUSION: A HUS needs to be large enough to reflect individuals' stated health preferences, which may lend support to the application of a minimally important difference for decision-making.