Item response theory analysis of the modified roland-morris disability questionnaire in a population-based study

Study Design. This is a secondary analysis of a cross-sectional population-based survey. Objective. Shorten the modified 23-item Roland (mRoland) scale using item response theory (IRT) methods and describe where in the functional disability range each scale is the most precise. Summary of Background Data. The Roland-Morris Disability Questionnaire is recommended for a functional disability outcome measure in patients with low back pain (LBP). One commonly used version is the Roland. It is unknown where in the functional disability range the Roland measures. Methods. One candidate individual with LBP in randomly selected households was interviewed, identifying 694 adults with chronic LBP. To justify the use of a unidimensional 2-parameter logistic IRT model, we performed both exploratory and confirmatory factor analysis. Results. Exploratory factor analysis revealed one dominant eigenvalue. Confirmatory factor analysis results indicate that the 1-factor model fit well. IRT analysis revealed variability in the slopes, in the range from 1.07 to 3.10. The marginal reliability, an IRT-based analog to coefficient α, was 0.88. The mRoland produces reliable scores (i.e., with a standard error <0.3) from 1.4 standard deviations below the mean to roughly 0.2 above the mean. Conclusion. The mRoland measures one construct. The mRoland seems to be an excellent tool for measuring just-below-average levels of functional disability. The mRoland measures high levels of functional disability with relatively poor reliability and may be more appropriate for a less-disabled population with LBP. We demonstrate that the mRoland can be shortened to 11 items with minimal loss of information. We show that there are different ways to go about selecting the set of 11 items that yield short forms with different strengths.