Concussion Assessment Across Several Clinical Batteries: Identifying the Components That Best Discriminate Injured Adolescents From Controls.

OBJECTIVE: To identify which sub-components of 4 clinical assessments optimize concussion diagnosis.

BACKGROUND: Multiple assessments are part of the clinical toolbox for diagnosing concussions in youth, including the Post-Concussion Symptom Inventory (PCSI), the visio-vestibular exam (VVE), the King-Devick (KD) assessment, and the Sport Concussion Assessment Tool (SCAT-5). Most of these assessments have sub-components that likely overlap in aspects of brain function they assess. Discerning the combination of sub-components that best discriminate concussed adolescents (cases) from uninjured controls would streamline concussion assessment.

DESIGN/METHODS: Participants, 12-18 years, were prospectively enrolled from August 1, 2017 to April 29, 2020 Controls (n = 189, 53% female) were recruited from a suburban high school with PCSI, VVE, KD and SCAT-5 assessments associated with their sport seasons. Cases (n = 213, 52% female) were recruited from a specialty care concussion program, with the same assessments performed ≤28 days from injury. We implemented a forward-selection sparse principal component (PC) regression procedure to group sub-components into interpretable PCs and identify the PCs best able to discriminate cases from controls while accounting for age, sex, and concussion history.

RESULTS: The AUC of the baseline model with age, sex, and concussion history was 62%. The PC that combined all 5 sub-components of PCSI and SCAT-5 symptom count and symptom severity provided the largest AUC increase (+10.6%) relative to baseline. Other PC factors representing (1) KD completion time, (2) Errors in BESS tandem and double-leg stances, and (C) horizontal/vertical saccades and vestibular-ocular reflex also improved model AUC relative to baseline by 5.6%, 4.7%, and 4.5%, respectively. In contrast, the SCAT5 immediate recall test and right/left monocular accommodation did little to uniquely contribute to discrimination (<1% gain in AUC). Overall, the best model included 5 PCs (AUC = 77%).

CONCLUSIONS: These data show overlapping features of clinical batteries, with symptoms providing the strongest discrimination, but unique features obtained from neurocognitive, vision, and vestibular testing.