An evaluation of the structural validity of the shoulder pain and disability index (SPADI) using the Rasch model
Christina Jerosch-Herold1, Rachel Chester2, Lee Shepstone3
1School of Health Sciences, Faculty of Medicine and Health Sciences, University of East Anglia, Norwich, Norfolk, NR4 7TJ, UK. c.jerosch-herold@uea.ac.uk.
Purpose:
The shoulder pain and disability index (SPADI) has been extensively evaluated for its psychometric properties using classical test theory (CTT). The purpose of this study was to evaluate its structural validity using Rasch model analysis.
Methods:
Responses to the SPADI from 1030 patients referred for physiotherapy with shoulder pain and enrolled in a prospective cohort study were available for Rasch model analysis. Overall fit, individual person and item fit, response format, dependence, unidimensionality, targeting, reliability and differential item functioning (DIF) were examined.
Results:
The SPADI pain subscale initially demonstrated a misfit due to DIF by age and gender. After iterative analysis it showed good fit to the Rasch model with acceptable targeting and unidimensionality (overall fit Chi-square statistic 57.2, p = 0.1; mean item fit residual 0.19 (1.5) and mean person fit residual 0.44 (1.1); person separation index (PSI) of 0.83. The disability subscale however shows significant misfit due to uniform DIF even after iterative analyses were used to explore different solutions to the sources of misfit (overall fit (Chi-square statistic 57.2, p = 0.1); mean item fit residual 0.54 (1.26) and mean person fit residual 0.38 (1.0); PSI 0.84).
Conclusions:
Rasch Model analysis of the SPADI has identified some strengths and limitations not previously observed using CTT methods. The SPADI should be treated as two separate subscales. The SPADI is a widely used outcome measure in clinical practice and research; however, the scores derived from it must be interpreted with caution. The pain subscale fits the Rasch model expectations well. The disability subscale does not fit the Rasch model and its current format does not meet the criteria for true interval-level measurement required for use as a primary endpoint in clinical trials. Clinicians should therefore exercise caution when interpreting score changes on the disability subscale and attempt to compare their scores to age- and sex-stratified data.
