From the fit statistics we cannot determine with certainty why 7% of our participants did not fit the Rasch model. It could be that they found the VAS scale difficult to understand and score; a qualitative investigation alongside this quantitative analysis could shed light on this. Taking these people out of the remaining analysis was important as their data led us to think the scale was not unidimensional; this would have been an incorrect conclusion as shown above. Three key findings arise from our study, which advance the field of research on the pain VAS. Firstly, our study showed that item difficulty of the pain VAS remained stable over a one-week period. This suggests the Pain VAS is interpreted in the same manner, irrespective of when it is completed and even when patients can see their previous scores. This lends support for the internal validity of the pain VAS. Secondly, we found that the pain VAS data fit the strict Rasch model, indicating it has internal validity. Thirdly, and importantly, the present analysis shows clearly that the pain VAS is an ordinal scale with a number of problems which makes its interpretation less straight forward: The pain VAS thresholds spread only over 1K to two logits. Such findings could occur if the sample is overly homogeneous. However, this was not the case here as table 1 and figures 1 and 2 showed that the narrow range occurred despite the use of 70% of the scale at HMPL-013 baseline and 98% of the scale at follow-up. Thus, the narrow range of thresholds is due to the lack of sensitivity of the VAS pain scale to EMD534085 distinguish between groups of people with different levels of pain. This finding is in contrast to commonly held beliefs that the VAS is sensitive in measuring pain. The range of logits found here is similar to the findings in the earlier WOMAC VAS scale study. Change in scores at the margins of the pain VAS, while gaining few raw score points, reflects considerable metric change. By contrast, moving across the middle of the pain VAS, gaining many raw score points, reflects little change on the metric. It follows from this that the magnitude of SRM��s depended on baseline pain VAS scores. For those with initial scores at the upper end or the lower end of the scale the SRMs were substantially higher on the metric than the ordinal equivalent. The pain VAS could therefore be said to be sensitive to change for those groups of patients.