Psychometric properties of Short Form-36 Health Survey, EuroQol 5-dimensions, and Hospital Anxiety and Depression Scale in patients with chronic pain

Abstract

Recent research has highlighted a need for the psychometric evaluation of instruments targeting core domains of the pain experience in chronic pain populations. In this study, the measurement properties of Short Form-36 Health Survey (SF-36),EuroQol 5-dimensions (EQ-5D) and Hospital Anxiety and Depression Scale (HADS) were analyzed within the item response-theory framework based on data from 35,908 patients. To assess the structural validity of these instruments, the empirical representations of several conceptually substantiated latent structures were compared in a cross-validation procedure. The most structurally sound representations were selected from each questionnaire and their internal consistency reliability computed as a summary of their precision. Finally, questionnaire scores were correlated with each other to evaluate their convergent and discriminant validity. Our results supported that SF-36 is an acceptable measure of 2 independent constructs of physical and mental health. By contrast, although the approach to summarize the health-related quality of life construct of EQ-5D as a unidimensional score was valid, its low reliability rendered practical model implementation of doubtful utility. Finally, rather than being separated into 2 subscales of anxiety and depression, HADS was a valid and reliable measure of overall emotional distress. In support of convergent and discriminant validity, correlations between questionnaires showed that theoretically similar traits were highly associated, whereas unrelated traits were not. Our models can be applied to score SF-36 and HADS in chronic pain patients, but we recommend against using the EQ-5D model due to its low reliability. These results are useful for researchers and clinicians involved in chronic pain populations because questionnaires' properties determine their discriminating ability in patient status assessment.

Figure 1.

Path diagrams representing the structural models used in this study. Factors are represented by circles, items by squares, and causal pathways by arrows. Unidimensional model: all items load on one single factor that accounts for their covariance. Correlated-traits model: item subsets load on separate factors that are correlated; accounts for between-item multidimensionality. Bifactor model: each item loads on 2 uncorrelated factors; one general for all items and one specific for item subgroups; accommodates within-item multidimensionality. Two-tier model: a bifactor model with multiple general factors, which can be correlated.

Figure 2.

Relationship between IRT scores and standardized test scores (left), and IRT scores and their standard errors (right). The points with the overlaid contour plots show individual observations and sample density calculated from the final model, with higher density in red areas, whereas LOESS lines depict trends per model. Larger IRT scores indicate better health for SF-36, lower HRQoL for EQ-5D, and higher emotional distress for HADS. BF, bifactor model; CT, correlation traits model; HADS, Hospital Anxiety and Depression Scale; HRQoL, health-related quality of life; IRT, item response theory; TTM, two-tier model; U, unidimensional model.

Figure 3.

Associations between the IRT scores from the final models (bold) and the conventional scores for SF-36, EQ-5D, and HADS. Larger IRT scores indicate better health for SF-36, lower HRQoL for EQ-5D, and higher emotional distress for HADS. Conventional scores were calculated based on Ware et al., 1993, the United Kingdom time trade-off value set, and as item summaries for SF-36, EQ-5D, and HADS, respectively. BP, bodily pain; GH, general health; HADS,Hospital Anxiety and Depression Scale; HRQoL, health-related quality of life; IRT, item response theory; MH, mental health; PCS and MCS, physical and mental component summary scores, respectively; PF, physical functioning; RE, role-emotional; RP, role-physical; SF, social functioning; VT, vitality. n = 31,050.