Is the pain visual analogue scale linear and responsive to change? An exploration using Rasch analysis

Abstract

Objectives: Pain visual analogue scales (VAS) are commonly used in clinical trials and are often treated as an interval level scale without evidence that this is appropriate. This paper examines the internal construct validity and responsiveness of the pain VAS using Rasch analysis.

Methods: Patients (n = 221, mean age 67, 58% female) with chronic stable joint pain (hip 40% or knee 60%) of mechanical origin waiting for joint replacement were included. Pain was scored on seven daily VASs. Rasch analysis was used to examine fit to the Rasch model. Responsiveness (Standardized Response Means, SRM) was examined on the raw ordinal data and the interval data generated from the Rasch analysis.

Results: Baseline pain VAS scores fitted the Rasch model, although 15 aberrant cases impacted on unidimensionality. There was some local dependency between items but this did not significantly affect the person estimates of pain. Daily pain (item difficulty) was stable, suggesting that single measures can be used. Overall, the SRMs derived from ordinal data overestimated the true responsiveness by 59%. Changes over time at the lower and higher end of the scale were represented by large jumps in interval equivalent data points; in the middle of the scale the reverse was seen.

Conclusions: The pain VAS is a valid tool for measuring pain at one point in time. However, the pain VAS does not behave linearly and SRMs vary along the trait of pain. Consequently, Minimum Clinically Important Differences using raw data, or change scores in general, are invalid as these will either under- or overestimate true change; raw pain VAS data should not be used as a primary outcome measure or to inform parametric-based Randomised Controlled Trial power calculations in research studies; and Rasch analysis should be used to convert ordinal data to interval data prior to data interpretation.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interests exist.

Figures

Figure 1. Data distribution baseline pain VAS scores (average over 7 days).

Figure 2. Data distribution follow-up pain VAS scores (average over 7 days).

Figure 3. Person Item Threshold distribution (Pre VAS data).

The graph displays the person-item threshold distribution map with the x-axes displaying location or difficulty of item thresholds (lower half) and location or level of pain reported on the VAS by participants (upper half). The y-axes display the frequencies of item thresholds (lower half) and participants (upper half). Thresholds of seven items are shown and it can be seen that the thresholds spread over 1½ logits only.

Figure 4. Item Response Curve for one VAS item.

The Item Response Curve displays the expected raw score on the y-axis and the interval transformed log score on the x-axis.

Figure 5. Standardised Response Means displayed by baseline raw pain VAS score.

Inclusion criteria into the study included a minimum score on a single pain VAS of 30; although on the seven daily VAS measures some scored below 30 mm, these numbers were small. Standardised Response Means (SRM) data for those with daily measurs below 30 mm have therefore been combined into one group.