Logical Inconsistencies in 3 Preference Elicitation Methods for EQ-5D Health States: A Study in the Brazilian Population.

BACKGROUND: . Logical inconsistency for health states preferences occurs when one logically worse health state, in terms of quality of life, is ranked higher than a logically better health state. OBJECTIVE: . This study explores the presence of inconsistent responses for the EQ-5D health states valuations in a Brazilian population survey. It compares the level of inconsistency in 3 preference-based methods: ranking, visual analog scale (VAS), and time tradeoff (TTO). The influence of EQ-5D health state descriptions is explored by examining the distance between states using a city-block metric as an indicator of proximity. Moreover, it examines the association between formal education and the presence of inconsistencies, as well as the effect of removing inconsistent respondents on the estimation of social value sets from TTO and VAS. METHODS: . Data came from a valuation study with 3362 literate individuals aged between 18 and 64 years living in urban areas of Minas Gerais state, Brazil. Logical inconsistency was assessed using the percentage of inconsistent respondents and inconsistency rate. A logistic model was estimated to assess the association between formal education and the logical inconsistency. Societal preferences were estimated excluding inconsistent respondents considering city-block metric. RESULTS: . The percentage of inconsistent respondents and inconsistency rate are similar for TTO and ranking and lower for VAS. The probability of being inconsistent is higher among less educated groups in ranking and TTO. Inconsistency decreases with distance for all 3 methods. The removal of inconsistent individuals by considering city-block distance improves TTO estimation of social value sets. CONCLUSION: . Findings suggest that removal of inconsistencies in TTO should consider city-block distance. For VAS, inconsistencies are not associated with formal education and do not affect social value set estimation.