Investigating the diagnostic accuracy of a paper-and-pencil and a computerized cognitive test battery for pediatric mild traumatic brain injury

May 29, 2022

Authors: Veronik Sicard, David D Stephenson, Danielle C Hergert, Andrew B Dodd, Cidney R Robertson-Benta, Sharvani Pabbathi Reddy, Keith Owen Yeates, Jason A Cromer, Timothy B Meier, Richard A Campbell, John P Phillips, Robert E Sapien, Andrew R Mayer

Journal: Neuropsychology

DOI: 10.1037/neu0000803

Year Published: 2022


This study assessed classification accuracy of paper-and-pencil and computerized cognitive batteries at subacute (SA; 1-11 days) and early chronic (EC; ∼4 months) phases of pediatric mild traumatic brain injury (pmTBI). Two statistical approaches focused on single-subject performance (individual task scores, total impairments) were used to maximize clinical utility.


Two hundred thirty-five pmTBI and 169 healthy controls (HC) participants aged 8-18 were enrolled, with a subset (190 pmTBI; 160 HC) returning for the EC visit. The paper-and-pencil battery included several neuropsychological tests selected from recommended common data elements, whereas computerized testing was performed with the Cogstate Brief Battery. Hierarchical logistic regressions (base model: Parental education and premorbid reading abilities; full model: Base model and cognitive testing variables) were used to examine sensitivity/specificity, with diagnosis as the dependent variable.


Number Sequencing and Cogstate One-Card Learning accuracy significantly predicted SA diagnosis (full model accuracy = 71.6%-71.7%, sensitivity = 80.6%-80.8%, specificity = 59.1%-59.6%), while only immediate recall was significant at EC visit (accuracy = 68.5%, sensitivity = 74.6%, specificity = 61.5%). Other measures (Letter Fluency, Cogstate Detection, and One-Card Learning accuracy) demonstrated higher proportions of impairment for pmTBI subacutely (pmTBI: 11.5%-19.8%; HC: 3.7%-6.1%) but did not improve classification accuracy. Evidence of multiple impairments across the entire testing battery significantly predicted diagnosis at both visits (full model accuracy = 66.2%-68.6%, sensitivity = 71.2%-78.9%, specificity = 54.3%-61.5%).


Current results suggest similar modest diagnostic accuracy for computerized and paper-and-pencil batteries across multiple pmTBI phases. Moreover, findings suggest the total number of impairments may be more clinically useful than any single test or cognitive domain in terms of diagnostic accuracy at both assessment points.

Back to Publications