Dimensionality and scaling properties of the Patient Categorisation Tool in patients with complex rehabilitation needs following acquired brain injury.

Abstract

To investigate the scaling properties of the Patient Categorisation Tool (PCAT) as an instrument to measure complexity of rehabilitation needs. Psychometric analysis in a multicentre cohort from the UK national clinical database. A total of 8,222 patents admitted for specialist inpatient rehabilitation following acquired brain injury. Dimensionality was explored using principal components analysis with Varimax rotation, followed by Rasch analysis on a random sample of n = 500. Principal components analysis identified 3 components explaining 50% of variance. The partial credit Rasch model was applied for the 17-item PCAT scale using a "super-items" methodology based on the principal components analysis results. Two out of 5 initially created super-items displayed signs of local dependency, which significantly affected the estimates. They were combined into a single super-item resulting in satisfactory model fit and unidimensionality. Differential item functioning (DIF) of 2 super-items was addressed by splitting between age groups (<65 and ≥ 65 years) to produce the best model fit (χ2/df = 54.72, p = 0.235) and reliability (Person Separation Index (PSI) = 0.79). Ordinal-to-interval conversion tables were produced. The PCAT has satisfied expectations of the unidimensional Rasch model in the current sample after minor modifications, and demonstrated acceptable reliability for individual assessment of rehabilitation complexity.