Testing for differential item functioning within the EQ-5D

We test for the presence of differential item functioning (DIF) in the EQ-5D health-related quality-of-life instrument, using data from a large clinical trial in acute stroke (ISRCTN 99414122). DIF occurs when subjects in different subsets of a sample respond differently to items in a measurement instrument, despite possessing the same latent traits. The data comprised 1462 patient records. We analyzed DIF specifically with respect to responses obtained from different geographical regions and responses obtained from proxies as opposed to the patients themselves. We mapped clinical outcome measures (scores from the modified Rankin Scale, the Barthel Index, and the Zung Depression scale) onto EQ-5D index scores and included dummy variables for proxy responses and for region of treatment (United Kingdom, Asia, rest of world). We predicted the level of problem severity reported on each of the EQ-5D's five constituent dimensions from the clinical measures and the dummy variables. For given clinical characteristics, proxies were more likely to report health problems than were the patients themselves, although the divergences were not sufficiently large to result in any significant difference in mean index scores between patient and proxy reports. However, the distributions of reported levels of problems for similar clinical states diverged significantly by region, and these translated into different index scores. The mean index score for UK responses was significantly higher than the mean index scores from Asia and the rest of the world.