Evaluating equivalency between response systems: application of the Rasch model to a 3-level and 5-level EQ-5D

Expansion of the EQ-5D health state classifier to 5 levels (EQ-5D-5L) has been proposed to improve discriminative and evaluative properties, but current preference-based algorithms were developed for a 3-level (EQ-5D-3L) structure. The objectives were to examine equivalency of meaning between 3L and 5L response systems, and to psychometrically derive a system of weights that facilitate conversion of 3L preference-based algorithms to a 5L system.

Rasch models were used to examine the equivalency of the 3L and 5L systems using 2 datasets where health status was assessed using the 3L and 5L: a Dutch study of primarily hypothetical health state assessments and a US-based multicenter study of 423 cancer patients. Category-specific mean values of latent person parameters (using maximum likelihood estimation) for the levels of the 3L and 5L systems were estimated.

Means on the latent continuum pertaining to level 3 in the 5L system and level 2 in the 3L (some problems) were similar for both datasets, suggesting equivalence of these levels. Extremes of the 5L response structure consistently broadened the measurement continuum. By anchoring "no problems" as 0 disutility, disutility weights from EQ-5D-3L were transformed into weights for EQ-5D-5L using ratios of logit distances between person means for 5L and 3L calibrated for each dimension using the Rasch model.

This study illustrates the rich potential for modern psychometric techniques both to examine equivalency when health status measures are modified as well as to inform preference-based measurement systems using existing value sets.