Item bias detection in the Hospital Anxiety and Depression Scale using structural equation modeling: comparison with other item bias detection methods

PURPOSE: Comparison of patient-reported outcomes may be invalidated by the occurrence of item bias, also known as differential item functioning. We show two ways of using structural equation modeling (SEM) to detect item bias: (1) multigroup SEM, which enables the detection of both uniform and nonuniform bias, and (2) multidimensional SEM, which enables the investigation of item bias with respect to several variables simultaneously. METHOD: Gender- and age-related bias in the items of the Hospital Anxiety and Depression Scale (HADS; Zigmond and Snaith in Acta Psychiatr Scand 67:361–370, 1983) from a sample of 1068 patients was investigated using the multigroup SEM approach and the multidimensional SEM approach. Results were compared to the results of the ordinal logistic regression, item response theory, and contingency tables methods reported by Cameron et al. (Qual Life Res 23:2883–2888, 2014). RESULTS: Both SEM approaches identified two items with gender-related bias and two items with age-related bias in the Anxiety subscale, and four items with age-related bias in the Depression subscale. Results from the SEM approaches generally agreed with the results of Cameron et al., although the SEM approaches identified more items as biased. CONCLUSION: SEM provides a flexible tool for the investigation of item bias in health-related questionnaires. Multidimensional SEM has practical and statistical advantages over multigroup SEM, and over other item bias detection methods, as it enables item bias detection with respect to multiple variables, of various measurement levels, and with more statistical power, ultimately providing more valid comparisons of patients’ well-being in both research and clinical practice. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s11136-016-1469-1) contains supplementary material, which is available to authorized users.