Quantifying the Value of Multidimensional Assessment Models for Acute Concussion: An Analysis of Data from the...

Many concussion assessment methods exist, but few studies quantify the performance of these methods to determine which can best assess acute concussion alone or in combination.

The objectives of this study were to evaluate: (1) selected concussion assessments for acute concussion assessment; (2) the utility of change scores for acute concussion assessment; and (3) concussion assessment capabilities when constrained to limited clinical data or objective clinical measures.

The 'acute concussion' group contained assessments from < 6 h post-injury (n = 560) and 24-48 h post-injury (n = 733). The 'normal performance' group contained assessments from baseline testing (n = 842) and unrestricted return to play (n = 707) timepoints. Univariate and multivariate logistic regression models were created separately for < 6- and 24- to 48-h timepoints. Models were evaluated on sensitivity, specificity, and area under the receiver operating characteristic curve.

Within the univariate analysis, Sport Concussion Assessment Tool symptom assessments had the highest combination of sensitivity, specificity, and area under the receiver operating characteristic curve, with values up to 0.93, 0.97, and 0.98, respectively. Full models had a sensitivity, specificity, and area under the receiver operating characteristic curve up to 0.94, 0.97, and 0.99, respectively, and outperformed all univariate models, raw score models, and objective models. Objective models were outperformed by all multivariate models and the univariate models containing only Sport Concussion Assessment Tool symptom assessments.

Results support the use of multidimensional assessment batteries over single instruments and suggest the importance of self-reported symptoms in acute concussion assessment. Balance assessments, however, may not provide additional benefit when symptom and neurocognitive assessments are available. Additionally, change scores provide some clinical utility over raw scores, but the difference may not be clinically meaningful.