Jerosch-Herold, Christina ORCID: https://orcid.org/0000-0003-0525-1282, Chester, Rachel
ORCID: https://orcid.org/0000-0003-1979-0682 and Shepstone, Lee
(2017)
Rasch model analysis gives new insights into the structural validity of the Quick-DASH in patients with musculoskeletal shoulder pain.
Journal of Orthopaedic and Sports Physical Therapy, 47 (9).
pp. 664-672.
ISSN 0190-6011
Preview |
PDF (Accepted manuscript)
- Accepted Version
Download (507kB) | Preview |
Abstract
Study Design: Cross-sectional secondary analysis of a prospective cohort study Background: The Quick-DASH is a widely used outcome measure which has been extensively evaluated using classical test theory (CTT). Rasch model analysis can identify strengths and weaknesses of rating scales which goes beyond CTT approaches. It uses a mathematical model to test the fit between the observed data and expected responses and converts ordinal-level scores into interval-level measurement. Objective: To test the structural validity of the Quick-DASH using Rasch analysis Methods: A prospective cohort study of 1030 patients with shoulder pain provided baseline data. Rasch analysis was conducted to i) assess how the Quick-DASH fits the Rasch model, ii) identify sources of misfit and iii) explore potential solutions to these. Results: There was evidence of multidimensionality and significant misfit to the Rasch model (χ²=331.04, p<0.001). Two items had disordered threshold responses with strong flooring effects. Response bias was detected in most items for age and gender. Rescoring resulted in ordered thresholds, however the 11-item scale still did not meet the expectations of the Rasch model. Conclusion: Rasch model analysis on the Quick-DASH has identified a number of problems which cannot be easily detected using traditional analyses. Whilst revisions to the Quick-DASH resulted in better fit, a ‘shoulder-specific’ version is not advocated at present. Caution needs to be exercised when interpreting results of the Quick-DASH outcome measure as it does not meet the criteria for interval level measurement and shows significant response bias by age and gender.
Downloads
Downloads per month over past year
Actions (login required)
![]() |
View Item |