IQ After Pediatric Concussion

Abstract

Objectives: This study investigated IQ scores in pediatric concussion (ie, mild traumatic brain injury) versus orthopedic injury.

Methods: Children (N = 866; aged 8-16.99 years) were recruited for 2 prospective cohort studies from emergency departments at children's hospitals (2 sites in the United States and 5 in Canada) ≤48 hours after sustaining a concussion or orthopedic injury. They completed IQ and performance validity testing postacutely (3-18 days postinjury; United States) or 3 months postinjury (Canada). Group differences in IQ scores were examined using 3 complementary statistical approaches (linear modeling, Bayesian, and multigroup factor analysis) in children performing above cutoffs on validity testing.

Results: Linear models showed small group differences in full-scale IQ (d [95% confidence interval] = 0.13 [0.00-0.26]) and matrix reasoning (0.16 [0.03-0.30]), but not in vocabulary scores. IQ scores were not related to previous concussion, acute clinical features, injury mechanism, a validated clinical risk score, pre- or postinjury symptom ratings, litigation, or symptomatic status at 1 month postinjury. Bayesian models provided moderate to very strong evidence against group differences in IQ scores (Bayes factor 0.02-0.23). Multigroup factor analysis further demonstrated strict measurement invariance, indicating group equivalence in factor structure of the IQ test and latent variable means.

Conclusions: Across multisite, prospective study cohorts, 3 complementary statistical models provided no evidence of clinically meaningful differences in IQ scores after pediatric concussion. Instead, overall results provided strong evidence against reduced intelligence in the first few weeks to months after pediatric concussion.

FIGURE 1

Flowchart summarizing how the final sample was derived for MIOS and the A-CAP study. MIOS participants who returned for the postacute assessment did not significantly differ from those who did not in terms of acute symptoms, sex, premorbid postconcussive symptom ratings, or age at time of injury.^, However, SES (based on the 2010 Census tract median family income in the ED) was higher among children who returned (n = 217) than who did not return (n = 98) for the MIOS postacute assessment, and a greater number of white and Asian American children returned for the postacute assessment than children of another race. Children in the MIOS study with completed WASI-II/SES data who scored below cutoffs on the MSVT (n = 18) did not differ from those who scored above the cutoffs in terms of age at injury, sex, race, or SES. For the A-CAP study, the children who returned (n = 719) at the 3-month follow-up did not differ significantly from those who did not return in age, sex, race, or parental education. Of those, 700 completed the WASI-II and also had completed SES data. However, 48 children with completed WASI-II/SES data scored below cutoffs on the MSVT and were excluded from final analyses; those who scored below cutoffs did not differ from those who scored above cutoffs in sex, SES, race, or injury mechanism, but were younger and more likely to have sustained the injury during nonsport-related activities.

FIGURE 2

Graphs illustrating IQ scores. Graphs illustrate postacute and 3 months postinjury full-scale IQ, matrix reasoning t, and vocabulary t scores for children with concussion and mild OI.

FIGURE 3

Graphic illustration of the factor model underlying the MIMIC invariance analyses. In the configural invariance model, the latent factor variance was fixed to 1.00 in both the concussion and mild OI groups for model identification. In the metric invariance model, factor loadings were fixed to equality to freely estimate the latent variance in the concussion group, σ²_IQ = 0.98 (95% CI, 0.63–1.32). Finally, in the scalar invariance model, the intercepts were fixed to equality to freely estimate the latent factor mean in the concussion group, η_IQ= −0.59 (95% CI, −1.17 to −0.01). Estimates (95% CI) for the configural invariance model are summarized for each group (denoted in subscripts) for each parameter, including standardized factor loadings (λ_j), unstandardized intercepts (τ_j), residual variances (ε_j), and MIMIC regression coefficients (β_j).