Abstract:
Context: Computerized neuropsychological testing is commonly utilized in the management of sport-related concussion. Few studies have addressed the sensitivity and specificity of these tools with a matched-control group design. It is important to provide statistical evidence regarding computerized testing and its ability to discern between concussed and non-concussed athletes. Objective: To determine the sensitivity and specificity of the ImPACT Test Battery in a collegiate athlete sample. Design: Crosssectional study from 2004-2008 sport seasons. Setting: Research Laboratory. Patients or Other Participants: Sixty-six collegiate athletes were classified as concussed or non-concussed (n = 33 per group; 25 males, 8 females each). Concussed group: age 19.73 + 1.4 years, education 13.24 + 1.2 years, nonconcussed group: age 19.03 + 1.7 years, education 12.76 + 1.2 years. The nonconcussed group reported no prior history of concussion and was matched to concussed participants by sex, sport, and position. Interventions: All participants completed baseline ImPACT testing prior to the start of their competitive season. All concussed participants were evaluated 24 hours postinjury. MANOVAs and post-hoc ANOVAs were utilized to identify group differences with α < .05. Discriminant function analysis was utilized to determine sensitivity and specificity with priors set at (.5, .5) and the external rule was applied for interpretation. The I score was calculated to determine how much better than chance a participant could be correctly classified as concussed or nonconcussed. Main Outcome Measures: ImPACT scores including verbal and visual memory, visual motor speed, reaction time, impulse control, and symptom scale score were analyzed. Results: Groups did not differ on age, years of education, handedness, history of special education, or diagnosis of learning disabilities (P > .05). Box’s test of equality of covariance matrices was significant, M = 128.83, F = 5.52, P < .001. Due to the sensitivity of the test and the fact we utilized equal sample sizes, we continued the analysis utilizing the linear rule. Significant differences were found between groups, Wilk’s = .744, F (6,59) = .744, P = .006,η2 = .256. Post-hoc ANOVAs revealed a significant difference between groups for symptom score, F (1,65) = 17.48, P < .001 and all other variables were non-significant (P > .05). Discriminant analysis correctly predicted overall group membership at 68.2%. Using predictive discriminate analysis, healthy participants were correctly predicted at 78.8% and concussed participants were correctly predicted at 57.6%. We found a strong I score (Z = 2.95, P < .05) which suggests 36.4% fewer classification errors would occur than if classification was done by chance. Conclusions: Overall, ImPACT had a 68.2% correct classification rate but significantly reduced classification errors. We advocate a battery of tests including a physical/ neurological examination in evaluating a concussion and making return to play decisions.