Tracking Spinal Cord Injury Functional Outcomes Across the Lifespan: Validation of Linking Coefficients

Abstract

ObjectiveValidation of linking coefficients to transform Pediatric Spinal Cord Injury Activity Measure (PEDI-SCI/AM) scores to adult Spinal Cord Injury-Functional Index (SCI-FI) scores. DesignThis cross-sectional study administered PEDI-SCI/AM and SCI-FI computerized adaptive tests (CATs) and short forms (SFs) to children with SCI and parents or caregivers. SettingHospitals, university, and rehabilitation institute. ParticipantsAbout 107 children with SCI and 96 parent or caregivers. InterventionsNot applicable. Main Outcome MeasuresLinking coefficients estimated SCI-FI (est-SCI-FI) scores from PEDI-SCI/AM scores for matched domains. Correlations between est-SCI-FI and actual SCI-FI scores were calculated. If correlations exceeded the criterion linking (0.866), the following analyses to compare est-SCI-FI and actual SCI-FI scores were conducted: paired t tests, intraclass correlation coefficients (ICCs 3, 1), percent of cases with absolute score differences at different thresholds. ResultsTwo matched domains, PEDI-SCI/AM Daily Routine/SCI-FI Self-Care and PEDI-SCI/AM General Mobility/SCI-FI Basic Mobility, met the linking criterion for both respondent-types (parent and child) and administration modes (CAT and SF). PEDI-SCI/AM Daily Routine and SCI-FI Fine Motor Function did not meet linking criterion for respondent type or mode. The linking criterion was met for wheelchair domains (child SF and CAT) and ambulation domains (child SF only). Significant differences between est-SCI-FI and actual SCI-FI scores were noted for all matched domains except Daily Routine/Self-Care (child SF only; parent SF and CAT). ICC values showed excellent agreement (range=0.75-0.89). Absolute differences between est-SCI-FI and actual SCI-FI scores were less than 1 standard deviation (except wheelchair CAT child). ConclusionsLinking coefficients applied to PEDI-SCI/AM scores can provide valid SCI-FI estimates that vary by domain, mode, and respondent type.