Background Several studies reported socio-economic (SE) gradients in physical function as assessed by HAQ. It is however not known whether this reflects true differences in physical function among socio-economic groups, or whether differential interpretation of HAQ items across SE groups biases item responses.
Objectives To investigate differential item functioning (DIF) across SE factors (age, gender, education and work status) when assessing physical function in RA patients across the different composite HAQ scores.
Methods Data from the Questionnaires in Standard Monitoring of Patients with Rheumatoid Arthritis (QUEST-RA), comprising 8,891 patients from 30 countries (mean age 55 years, 81% females) were used. Physical function in daily activities was assessed using a set of 30 individual items from the different HAQ versions (Stanford HAQ, HAQ-II, modified HAQ (MHAQ) and multi-dimensional HAQ (MDHAQ)). SE factors comprised age, gender, education (years), and work status (yes or no).
DIF was investigated using item response theory models implemented in a latent variable modelling framework, with models equivalent to ordinal logistic regressions having answers on each HAQ as outcome and SE factors as well as the latent trait function as predictors. Scores for 4 composite HAQs were recalculated and adjusted for DIF by methods similar as above, but simultaneous modelling all previously identified items exhibiting DIF. To assess the impact of DIF on associations between SE factors and composite HAQ scores, multilevel mixed effect linear regression models were estimated with individuals nested in country. Models were adjusted for SE factors, disease activity score in 28 joints (DAS28)) and a comorbidity index (Rheumatic diseases comorbidity index (RDCI)). Changes in the strength of association between SE-factors and HAQ or for DIF-adjusted HAQ were investigated by comparing β-coefficient of the SE factors in the models.
Results Relevant DIF (defined as OR >1.1 or <0.90, (Δ≥10%)) in HAQ items was found for the SE factors age (n=13 items), gender (n=26 items) and work status (n=20 items), whereas education influenced only few items (n=7) (table) (table). Adjustment of composite HAQs for DIF resulted in an increase of HAQ scores, indicating influence of individual item DIF on the composite scores (table). In regression models all SE factors remained significantly related to all composite HAQ before or after adjusting for DIF. Changes in coefficients of regression models were overall negligible, except for gender (all HAQ versions). β-coefficients for gender increased by up to 0.07 (70% coefficient change), indicating several items have for women a different meaning, independent of their underlying level of functional limitation.
Conclusions Relevant measurement bias across SE factors was found for a number of individual HAQ items. In composite HAQ scores, some evidence for measurement issues remains. Although not having a major impact in terms of measurement of physical function, there is some need for caution when comparing HAQ between males and females.
Disclosure of Interest None declared