Study design: A cross-sectional survey using patient questionnaires was conducted. OBJECTIVE To assess whether the Roland Disability Questionnaire satisfies the Rasch model including unidimensionality and item separation.
Summary of background data: The Roland Disability Questionnaire, the most widely used patient-assessed measure of health outcome for back pain, has undergone several evaluations for its measurement properties including reliability, validity, and responsiveness. However, there is no published work relating to the underlying dimensionality of the instrument and the extent to which individual items contribute to the construct of physical disability resulting from low back pain.
Methods: Patients entering a randomized controlled trial of exercise, manipulation, and usual management for back pain completed a questionnaire that included the Roland Disability Questionnaire. The Winsteps program was used to assess whether the Roland Disability Questionnaire fits the Rasch model. Item fit was assessed using the Infit and Outfit statistics.
Results: The Roland Disability Questionnaire was completed by 1008 (90%) of the patients taking part in the trial. Most of the items in the Roland Disability Questionnaire contribute to a single underlying construct. However, four items had poor Outfit statistics, suggesting that they do not contribute sufficiently to the scale hierarchy. Several items positioned around the middle of the hierarchy are not sufficiently distinct in terms of difficulty. There were very few items positioned at the extremes of the hierarchy.
Conclusions: The Roland Disability Questionnaire largely satisfies the Rasch model for unidimensionality. However, the instrument could be improved through the removal of poorly fitting items and the addition of items at the upper and lower points of the scale hierarchy. The distribution of Roland Disability Questionnaire scores should be carefully considered before statistical testing is undertaken. Rasch transformed scores can be used to deal with deficiencies in the scale hierarchy.