Development of a Short Form Telephone-based Interview for Caregivers to Monitor Patient Outcome After Severe Brain Injury

Research Objectives: To develop a psychometrically-sound caregiver administered short form interview to assess neurobehavioral function in persons with severe disability after brain injury.

Design: 117 caregivers cross-sectionally completed the standard long form of the telephone-administered Coma Recovery Scale–Revised (CRS-R) and Confusion Assessment Protocol Cognitive Impairment Subscale and Galveston Orientation and Amnesia Test (CAP-Cog/GOAT).

Setting: 6 inpatient rehabilitation hospitals.

Participants: Caregivers of persons with severe brain injury (n=117 caregivers, mean [SD] age=55 [13]; 80% female); (n=117 TBI patients; mean [SD] age=34 [18]; 25% female).

Interventions: N/A

Main Outcome Measures: Confirmatory and exploratory factor analysis was performed to evaluate the underlying latent constructs of interest for the CRS-R and the CAP-Cog/GOAT. Construct validity using unidimensional confirmatory factor analyses (CFA) with the robust weighted least square estimation method was conducted. Further, exploratory bi-factor analysis examined the percent of Explained Common Variance (ECV) and Omega Hierarchical (omegaH: percentage of the variance of the total score that is contributed to the general factor). We then applied the Rasch Partial Credit Model and Rating Scale Model to fit the data. Last, we conducted Differential Item Functioning (DIF) analysis for demographic and clinical variables.

Results: The confirmatory factor analysis did not fit the data for a single factor; however, using exploratory factor analysis, the first factor explained 60% of the variance and the ratio of the 1st and 2nd eigenvalues was 4.74. Further, the explained common variance was > 0.70 and omegaH was > 0.80, all evidence of a strong single factor. Rasch analysis based on a rating scale model with threshold parameters set equally across the 17 items represented on the long form showed no disordered threshold parameters. Two misfit items and 3 local dependent items were sequentially removed. The final short form rating scale model had 12 items (7 items from CRS-R and 5 items from CAP-Cog/GOAT) giving 85% of the variance explained by the Rasch model, with person score reliability of 0.90, with no misfit items. The short form had < 5% with ceiling and floor effects. 76% of the sample fit the Rasch model. Last, all demographic and clinical variables tested indicated small DIF effects.

Conclusions: The final rating scale short form model is psychometrically sound with 12 items. The most difficult item gauges visual memory and the easiest item gauges command-following. The next phase of this work will assess clinical applicability.

Author(s) Disclosures: None