Measuring Neurobehavioral Function in Pediatrics: A Rasch Analysis of the Coma Recovery Scale - Revised

Research Objectives: To investigate preliminary evidence for the structural validity and reliability of the Coma Recovery Scale-Revised (CRS-R) and CRS-Pediatric (CRS-P) assessments) administered to pediatric patients with disorders of consciousness.

Design: Retrospective analysis of the CRS-R and CRS-P using the Rasch measurement model for repeated measures.

Setting: Inpatient rehabilitation at a private children’s hospital.

Participants: Children (1-18 years old) with disorders of consciousness following severe brain injury.

Interventions: NA

Main Outcome Measures: The CRS-R is the reference standard for assessing neurobehavioral function in patients with disorders of consciousness >5 years old. The CRS-P was adapted for children who are < 5 years old; rating scale categories are similar although the stimuli used to elicit a neurobehavioral response differ. For this study, we combined data from both assessments.

Results: The average age of the 47 participants was 103 months (SD: 60 months) and on average they were 60 days post-injury at first CRS-R/P assessment. Each participant had >1 CRS-R/P record for a total of 1249 CRS-R records. After randomizing first and last records to a calibration/validation sample, we used the Rasch model to generate anchored item and rating scale measures along an equal-interval scale ranging from -7.10 to 6.65 logits. In order from highest to lowest neurobehavioral function, the item hierarchy (logits) was Communication (2.71), Visual (0.07), Auditory (0.05), Verbal/Oromotor (-0.61), Motor (-0.93) and Arousal (-1.29). individual rating scale categories spanned -6.65 to 5.63 logits. All items showed acceptable model fit. Principal components analysis revealed eigenvalue 1.59. The separation and person reliability indices for the CRS-R were 2.11 and 0.85, respectively. For all items, the most difficult rating scale categories were used < 5% of all observations. Putting all data into the model, we found that 81% of CRS-R records fit the model and mean person measure was -1.39 logits.

Conclusions: Preliminary Rasch analysis of the CRS-R/P in pediatric patients suggests that the items comprise a unidimensional hierarchy of neurobehavioral function and are sufficiently reliable for group-level decisions. Because the more challenging rating scale categories were used less frequently, they may be too difficult for young children to achieve. This may also reflect common clinical practice of discontinuing the CRS-R/P once the child achieves higher scores. However, because some categories were used fewer than 10 times in the calibration dataset, more robust analyses are necessary to determine structural validity and reliability of the CRS-R/P.

Author(s) Disclosures: NA