Scoring the EQ-HWB-S: can we do it without value sets? A non-parametric item response theory analysis

Abstract

BackgroundOnly one pilot value set (UK) is currently available for the EQ Health and Wellbeing Instrument short version (EQ-HWB-S). As an alternative to preference-weighted scoring, we examined whether a level summary score (LSS) is appropriate for the EQ-HWB-S using Mokken scaling analyses.MethodsData from patients, carers and the general population collected during the developmental phase of the EQ-HWB-S in Australia, US and UK were used, noting 3 of 9 items have since undergone revision. EQ-HWB-S data fit was examined using R package Mokken scaling’s monotone homogeneity model, utilizing the automated item selection procedure (AISP) as well as Loevinger’s scaling coefficients for items and the scale (HS). Manifest monotonicity was assessed by examining whether the cumulative probability for responses at or above each response level did not decrease across the summary score.ResultsEQ-HWB-S data were available for 3340 respondents: US = 903, Australia = 514 and UK = 1923. Mean age was 50 ± 18 and 1841 (55%) were female. AISP placed all 9 items of the EQ-HWB-S on a single scale when the lower bound was set to < 0.448. Strong scalability (HS = 0.561) was found for the EQ-HWB-S as a single scale. Stronger scales were formed by separating the psychosocial items (n = 6, HS = 0.683) and physical sensation items (n = 3, HS = 0.713). No violations of monotonicity were found except for the items mobility and daily activities for the subgroups with long-term conditions and UK subjects, respectively.DiscussionAs EQ-HWB-S items formed a strong scale and subscales based on Mokken analysis, LSS is a promising weighting-free approach to scoring.