OBJECTIVES: Value sets for both the EQ-5D-3L (‘3L’) and the EQ-5D-5L (‘5L’) are now available in the US. Experience in other countries has shown that the choice of measure, e.g. using the 3L or 5L, has implications for the cost-effectiveness of interventions in health care. The objective of this study was to compare and contrast United States value sets for the EQ-5D-3L (Shaw et al 2005) and EQ-5D-5L (Pickard et al, 2019) scoring functions.

METHODS: The 3L and 5L value sets were characterized in terms of distribution normality, percentage of worse-than-dead (WTD) health states (utility < 0), and range of scale. The value sets were also applied to respondents from 2 cross-sectional datasets: the multi-country study upon which the crosswalk value set was based (n=3919), and US face-to-face and online valuation (n=3152) studies to compare discriminative ability across a range of respondent characteristics. This was assessed by ability of value sets to distinguish between known groups, e.g., EQ-5D-5L dimension-levels, using F statistic ratios, where values greater than 1 indicate greater discriminative ability.

RESULTS: The US 3L value set had a shorter range of scale than the 5L ([3L] -0.109 to 1 vs. [5L] -0.573 to 1). Further, the 3L had a bimodal distribution whereas the 5L has a unimodal, symmetric distribution. Ten of 243 (4.1%) 3L health states and 620 of 3,125 (19.8%) 5L health states were WTD. In both the US valuation and crosswalk data, the 5L value set was more statistically efficient in discriminating among levels of health using various criteria based on health problems (F-ratios 1.15 to 3.67).

CONCLUSIONS: Results suggest that the differences between the 3L and 5L value sets are likely to have implications for cost-effectiveness and statistical efficiency which may favor the 5L, but more research using a wide array of empirical datasets and simulation is needed.