1 Department of Sports Science and Clinical Biomechanics, Det Sundhedsvidenskabelige Fakultet, SDU2 Clinical Biomechanics, Department of Sports Science and Clinical Biomechanics, Det Sundhedsvidenskabelige Fakultet, SDU3 Department of Sports Science and Clinical Biomechanics, Det Sundhedsvidenskabelige Fakultet, SDU
Title Rasch scaling of the Oswestry Disability Index and the Roland-Morris Disability Index Authors & Affiliations: Henrik Hein Lauridsen1, Jan Hartvigsen1,2 1. Research Unit for Clinical Biomechanics, Institute of Sports Science and Clinical Biomechanics, University of Southern Denmark, Clinical Locomotion Network, Campusvej 55, DK-5230 Odense M, Denmark 2. Nordic Institute of Chiropractic and Clinical Biomechanics, Clinical Locomotion Network, Forskerparken 10A, 5230 Odense M, Denmark Background Rasch analysis is a modern psychometric approach with the ability to turn instruments scored on an ordinal scale into interval scaling in addition to optimising the fit of instrument items to the target population. In low back pain research the two most commonly used and well-validated questionnaires to assess functional status in patients with low back pain are the Roland-Morris Disability Questionnaire (RMQ) and the Oswestry Disability Index (ODI), however, only few studies have tested these questionnaires using Rasch analysis. This study used Rasch scaling to test the construct validity of the Danish versions of the RMQ (23-item Patrick version) and the ODI (version 2.1a) in a heterogeneous population of low back pain patients. Methods A total of 232 patients with low back pain and/or leg pain were recruited from an out-patient spinal clinic and 7 chiropractic practices and assessed using the RMQ and the ODI. The Dichotomous Model was applied to the RMQ and the unrestricted Partial Credit Model was used for the ODI. A series of Rasch procedures were carried out including overall fit, the person separation index (PSI), item fit, threshold ordering and differential item functioning (DIF). In addition local independence, unidimensionality and targeting were examined. The Rasch analysis was carried out using RUMM2030 and an acceptable overall fit of the data to the model was set as P>0.01 (item-trait interaction, χ2 probability). Items were considered misfitting if fit residuals exceeded ∓ 2.5 or χ2 probability was < 0.01. DIF was considered significant if the Bonferroni corrected χ2 probability was < 0.001. Results Sample characteristics revealed a mean age of 44 years, that 51% were females and 43% had pain in the LB and the leg. 55% were seen in primary care. ODI The ODI showed adequate overall fit to the Rasch Model (P > 0.01) with a PSI of 0.86. There were no misfitting items, however items 2,4,6,8-10 showed disordered thresholds. Uniform DIF was seen with where patients presented (primary or secondary sector) and duration. Local dependence was found between items 4 and 6. A paired t-test revealed marginal unidimensionality (lower bound 95% CI: 0.07). Targeting showed that the patients had less disability (mean: -1.34) compared to item difficulty. RMQ The RMQ showed poor overall fit to the model (P < 0.01) despite an adequate PSI of 0.83. 6 items misfitted the model (3,5,17,18,21 and 23). Item 9 showed uniform DIF with duration and 4 pairs of items were locally dependent. A paired t-test showed marginal unidimensional (lower bound 95% CI: 0.08). Targeting showed several gaps and that the patients had more disability compared to item difficulty (mean: 0.28). Conclusion The Danish version of the ODI largely satisfies the Rasch model whereas the RMQ did not. Both instruments showed satisfactory PSI indicating good discrimination at two levels of ability. The ODI and RMQ could be improved through the removal of poorly fitting and locally dependent items. Future studies should address the poor targeting by adding items at the extremes to the RMQ and easy items to the ODI.