OBJECTIVE: To assess the predictive capability of the postconcussion symptom scale (PCSS) of the sport concussion assessment tool (SCAT) III to differentiate concussed and nonconcussed adolescents. DESIGN: Retrospective.Tertiary. PARTICIPANTS: Sixty-nine concussed (15.2 ± 1.6 years old) and 55 control (14.4 ± 1.7 years old) adolescents. INDEPENDENT VARIABLES: Postconcussion symptom scale. MAIN OUTCOME MEASURE: Two-proportion z-test determined differences in symptom endorsement between groups. To assess the predictive power of the PCSS, we trained an ensemble classifier composed of a forest of 1000 decision trees to classify subjects as concussed, or not concussed, based on PCSS responses. The initial classifier was trained on all 22-concussion symptoms addressed in the PCSS, whereas the second classifier removed concussion symptoms that were not statistically significant between groups. RESULTS: Concussion symptoms common between groups were trouble falling asleep, more emotional, irritability, sadness, and anxious. After removal, analysis of the second classifier indicated that the 5 leading feature rankings of symptoms were headache, head pressure, light sensitivity, noise sensitivity, and "don't feel right," which accounted for 52% of the variance between groups. CONCLUSIONS: Collectively, self-reported symptoms through the PCSS can differentiate concussed and nonconcussed adolescents. However, predictability for adolescent patients may be improved by removing emotional and sleep domain symptoms.
Available at: http://works.bepress.com/kevin-shoemaker/5/