Another Look at the PART-O Using the Traumatic Brain Injury Model Systems National Database: Scoring to Optimize Psychometrics

Abstract

ObjectiveTo integrate previous approaches to scoring the Participation Assessment with Recombined Tools-Objective (PART-O) in a unidimensional scale. DesignRetrospective analysis of PART-O data from the Traumatic Brain Injury Model Systems. SettingCommunity. ParticipantsData from individuals (N=469) selected randomly from participants who completed 1-year follow-up in the Traumatic Brain Injury Model Systems were used in Rasch model development. The model was subsequently tested on data from additional random samples of similar size at 1-, 2-, 5-, 10-, and >15-year follow-ups. InterventionsNot applicable. Main Outcome MeasurePART-O. ResultsAfter combining items for productivity and social interaction, the initial analysis at 1-year follow-up indicated relatively good fit to the Rasch model (person reliability=.80) but also suggested item misfit and that the 0-to-5 scale used for most items did not consistently show clear separation between rating levels. Reducing item rating scales to 3 levels (except combined and dichotomous items) resolved these issues and demonstrated good item level discrimination, fit, and person reliability (.81), with no evidence of multidimensionality. These results replicated in analyses at each additional follow-up period. ConclusionsModifications to item scoring for the PART-O resulted in a unidimensional parametric equivalent measure that addresses previous concerns about competing item relations, and it fit the Rasch model consistently across follow-up periods. The person-item map shows a progression toward greater community participation from solitary and dyadic activities, such as leaving the house and having a friend through social and productivity activities, to group activities with others who share interests or beliefs.