A Rasch analysis of the Brief Pain Inventory Interference subscale reveals three dimensions and an age bias

David M Walton; Tyler Beattie; Joseph Putos; Joy C MacDermid

doi:10.1016/j.jclinepi.2015.10.022

A Rasch analysis of the Brief Pain Inventory Interference subscale reveals three dimensions and an age bias

J Clin Epidemiol. 2016 Jun:74:218-26. doi: 10.1016/j.jclinepi.2015.10.022. Epub 2016 Jan 6.

Authors

David M Walton¹, Tyler Beattie², Joseph Putos², Joy C MacDermid³

Affiliations

¹ School of Physical Therapy, Western University, Rm. EC1443, 1201 Western Rd, London, Ontario, Canada N6G 1H1. Electronic address: dwalton5@uwo.ca.
² School of Physical Therapy, Western University, Rm. EC1443, 1201 Western Rd, London, Ontario, Canada N6G 1H1.
³ School of Physical Therapy, McMaster University, Rm. IAHS403, 1400 Main St. W., Hamilton, Ontario, Canada L8S 1C7.

PMID: 26769256
DOI: 10.1016/j.jclinepi.2015.10.022

Abstract

Objectives: The Brief Pain Inventory is composed of two quantifiable scales: pain severity and pain interference. The reported factor structure of the interference subscale is not consistent in the extant literature, with no clear choice between a single- or two-factor structure. Here, we report on the results of Rasch-based analysis of the interference subscale using a large population-based ambulatory patient database (the Quebec Pain Registry).

Study design: Observational cohort.

Results: A total of 1,000 responses were randomly drawn from a total database of 5,654 for this analysis. Both the original 7-item and an expanded 10-item version (Tyler 2002) of the interference subscale were evaluated. Rasch analysis revealed significant misfit of both versions of the scale, with the original 7-item version outperforming the expanded 10-item version. Analysis of dimensionality revealed that both versions showed improved model fit when considered two subscales (affective and physical interference) with the item on sleep interference removed or considered separately. Additionally, significant uniform differential item functioning was identified for 6 of the 7 original items when the sample was stratified by age above or below 55 years. The interference subscale achieved adequate model fit when considered as two separate subscales with age as a mediator of response, while interpreting the sleep interference item separately. A transformation matrix revealed that in all cases, ordinal-level change at the extreme ends of the scale appears to be more meaningful than does a similar change at the midpoints.

Conclusions: The Interference subscale of the BPI should be interpreted as two separate subscales (Affective Interference, Physical Interference) with the sleep item removed or interpreted separately for optimal fit to the Rasch model. Implications for research and clinical use are discussed.

Keywords: Factor structure; Interval-level scoring; Measurement; Pain; Psychometrics; Rasch; Validitiy.

Publication types

Observational Study
Research Support, Non-U.S. Gov't

MeSH terms

Adolescent
Adult
Age Distribution
Aged
Aged, 80 and over
Bias
Cohort Studies
Female
Health Surveys / methods*
Health Surveys / statistics & numerical data*
Humans
Male
Middle Aged
Models, Statistical*
Pain / epidemiology*
Psychometrics
Quality of Life
Quebec / epidemiology
Reproducibility of Results
Young Adult

Grants and funding

MOP – 137148/Canadian Institutes of Health Research /International