Locally derived transition probabilities for EQ-5D crosswalk in a Brazilian population: an alternative to European-based mapping algorithms

Measuring health-related quality of life using Quality-Adjusted Life Years (QALYs) is central for health technology assessment. While Brazil has a validated EQ-5D-3 L value set, no nationally representative EQ-5D-5 L value set is yet available. Mapping algorithms such as the nonparametric crosswalk developed by van Hout et al. (2012) address this gap by using empirical response-transition probabilities between 5 L and 3 L items; however, these probabilities were derived from European populations and may not reflect Brazilian response patterns. To derive locally based transition probabilities from a Brazilian sample and use them within the nonparametric crosswalk framework to generate EQ-5D-5 L utility estimates, and to compare these with estimates obtained using the original European-derived transition probabilities. A cross-sectional study was conducted with 413 Brazilian participants (310 with myasthenia gravis; 103 from a social intervention program [1]). Both EQ-5D-3 L and EQ-5D-5 L were administered during the same assessment session, with the 3 L preceding the 5 L. Empirical conditional response probabilities of EQ-5D-3 L levels given each EQ-5D-5 L level were derived per dimension and used to estimate utilities for all 3,125 EQ-5D-5 L health states, weighted by the validated Brazilian EQ-5D-3 L tariff. Uncertainty was quantified via non-parametric bootstrap resampling of participants. Results were compared with estimates from the van Hout algorithm using European transition probabilities applied with the same Brazilian tariff. The locally derived model produced a mean utility of 0.375 (SD 0.162) across 3,125 EQ-5D-5 L states, compared with 0.385 (SD 0.147) for the European-based model. The Brazilian model classified more states as worse than death (1.73%; n = 54) than the European model (0.45%; n = 14). Lin’s concordance correlation coefficient (CCC) was 0.984 (95% CI: 0.774–0.984), and Spearman’s ρ was 0.991 ( p < 0.001), indicating excellent overall agreement. Divergence was concentrated at the extreme end of the severity spectrum, where the Brazilian model assigned utilities up to 0.079 points lower than the European model. In this sample, transition probabilities differed from European estimates, particularly for severe health states; whether this reflects population-specific response patterns or sample-related factors could not be established. The locally derived crosswalk may offer an interim, sample-specific option for cost-utility analyses in similar populations, pending validation in independent Brazilian datasets and, ultimately, a national EQ-5D-5 L value set.