A comparison of scaling methods for psychophysical ratings of acute musculoskeletal pain

摘要

The purpose of this study was to develop a psychophysically and psychometrically valid pain measurement strategy. Current pain measurement methods use a linear psychometric model known as classical test theory (CTT). A contemporary nonlinear psychometric model, item response theory (IRT), has not been used to develop a pain measurement strategy. Subjects who were experiencing acute musculoskeletal pain were asked to perform a psychophysical matching task to rate their perceptions of pain relative to qualitative verbal descriptor categories in three pain dimensions: sensory-intensity; cognitive-evaluative; and affective-emotional. The CTT and IRT psychometric models were used to scale the verbal category ratings within pain dimensions. CTT statistics were estimated using confirmatory factor analysis to fit a common factor model for each pain dimension. IRT statistics were estimated using the graded response model for each pain dimension. The psychometric models were compared with respect to standardized residuals (model fit), standard errors of measurement (model precision), and test characteristic curves (model bias). Across pain dimensions, the IRT model produced fewer significant (p < 0.01) standardized residuals than did the CTT model (51 vs. 96, respectively). The IRT model also produce fewer significant standardized residual subject scores than the CTT model for the sensory-intensity (10 vs. 13, respectively), cognitive-evaluative (3 vs. 6), and emotional-affective (0 vs. 9) dimensions. The IRT model produced smaller standard errors of measurement than did the CTT model for the majority of the scale range. Test characteristic curves showed that the CTT model produced considerable bias for much of the scale range, particularly for the cognitive-evaluative and emotional-affective dimensions. Several additional advantages of the IRT model over the CTT model are discussed, including analyses for the fit of individual subjects' responses and parameter invariance. Because of superior model fit in two of the three pain dimensions, superior score precision, and less scale bias, the IRT psychometric model was determined to be a superior strategy for scaling psychophysical ratings of acute musculoskeletal pain. Future investigations using pain rating scales should apply IRT measurement models in order to attain better scores.