Measuring disparities: bias in the Short Form-36v2 among Spanish-speaking medical patients.

Background: Many national surveys have found substantial differences in self-reported overall health between Spanish-speaking Hispanics and other racial/ethnic groups. However, because cultural and language differences may create measurement bias, it is unclear whether observed differences in self-reported overall health reflect true differences in health.
Objectives: This study uses a cross-sectional survey to investigate psychometric properties of the Short Form-36v2 for subjects across 4 racial/ethnic and language groups. Multigroup latent variable modeling was used to test increasingly stringent criteria for measurement equivalence.
Subjects: Our sample (N=1281) included 383 non-Hispanic whites, 368 non-Hispanic blacks, 206 Hispanics interviewed in English, and 324 Hispanics interviewed in Spanish recruited from outpatient medical clinics in 2 large urban areas.
Results: We found weak factorial invariance across the 4 groups. However, there was no evidence for strong factorial invariance. The overall fit of the model was substantially worse (change in Comparative Fit Index >0.02, root mean square error of approximation change >0.003) after requiring equal intercepts across all groups. Further comparisons established that the equality constraints on the intercepts for Spanish-speaking Hispanics were responsible for the decrement to model fit.
Conclusions: Observed differences between SF-36v2 scores for Spanish-speaking Hispanics are systematically biased relative to the other 3 groups. The lack of strong invariance suggests the need for caution when comparing SF-36v2 mean scores of Spanish-speaking Hispanics with those of other groups. However, measurement equivalence testing for this study supports correlational or multivariate latent variable analyses of SF-36v2 responses across all the 4 subgroups, as these analyses require only weak factorial invariance.