Objective: The objectives of this study were: 1) to determine whether expanding the number of levels (ie, response categories) on the standard 3 level EQ-5D (EQ-5D-3L) to 5-levels (EQ-5D-5L) would improve the descriptive richness and ability of the measure to discriminate among different levels of health, and 2) to examine the psychometric properties of each EQ-5D version in patients with cancer.
Methods: U.S.-based cancer patients self-assessed their health using EQ-5D-3L and EQ-5D-5L. These versions were compared in terms of ceiling effects, convergent validity based on correlations with ECOG performance status and FACT-G, discriminative ability using Rasch analysis, and informational richness using Shannon's Evenness Index (J').
Results: A ceiling effect was observed among a greater proportion of respondents to EQ-5D-3L, n=74/424 (17%), compared with EQ-5D-5L, n=47/424 (11%). Within the midlevel of EQ-5D-3L (some problems), substantial partitioning of the sample into the 3 nonextreme levels of the EQ-5D-5L was observed across dimensions. EQ-5D-5L demonstrated a trend towards slightly stronger correlations with ECOG performance status compared with EQ-5D-3L for all dimensions of health, ie, rs (5L/3L): rmobility=0.38/0.31; rself-care=0.35/0.31; rusual activities=0.55/0.47; rpain/discomfort=0.43/0.37; ranxiety/depression=0.23/0.16; rcrude summary score=0.56/0.49. EQ-5D-5L demonstrated a greater relative efficiency and ability to discriminate different levels of health. Informational richness and evenness of EQ-5D-5L was slightly higher (J'5L=0.75) than EQ-5D-3L (J'3L=0.69).
Conclusion: Evidence supported the validity of both EQ-5D-3L and EQ-5D-5L in cancer. However, results suggest a 5-level classifier system has less ceiling effect and greater discriminative ability with potentially more power to detect differences between groups compared with EQ-5D-3L.