Equivalence in the Health Assessment Questionnaire (HAQ) across socio-demographic determinants: Analyses within QUEST-RA

Objectives: To investigate potential bias in scores of the Health Assessment Questionnaire (HAQ) related to socio-demographic (SD) background of patients with rheumatoid arthritis (RA). Methods: Data from the Quantitative Standard Monitoring of Rheumatoid Arthritis study (QUEST-RA), comprising 9022 patients were analysed. Physical function was assessed through 30 items of four HAQ versions: the HAQ-Disability scale, HAQ-II, modified HAQ and multi-dimensional HAQ (MD-HAQ). DIF was investigated using item response theory models implemented in a latent variable modelling framework. Models were equivalent to ordinal logistic regression models with HAQ score (item level) as outcome, the latent trait 'physical function' and individual SD factors (age, gender, education, and employment status) as predictors. Next, scores of composite HAQs were adjusted for DIF. To assess the impact of DIF on associations between SD factors and HAQs, multilevel mixed-effect linear regression models with individuals nested in country were estimated with DIF-adjusted or unadjusted HAQ as outcome. Results: Relevant DIF (OR > 1.1 or < 0.90) was found in several HAQ items primarily for age, gender and work status. Adjustment of composite HAQs for DIF resulted in small increases (Delta 0.02-0.07); MD-HAQ best compensated for bias related to SD factors (Delta 0.02). In regressions, all SD factors remained significantly related to DIF-adjusted HAQs, with differences in coefficients largest for gender (Delta 0.02-0.07) but overall negligible. Conclusions: SD factors produce response bias in individual HAQ items but have little impact on composite HAQs. When interpreting HAQ across SD factors, MD-HAQ is preferred, but caution remains when comparing function across gender. (C) 2018 Elsevier Inc. All rights reserved