Neurosensory Screening and Symptom Provocation in Pediatric Mild Traumatic Brain Injury

Abstract

Objective:

To evaluate diagnostic/prognostic implications of neurosensory testing during the sub-acute stage in patients with pediatric mild traumatic brain injury (pmTBI).

Setting:

Recruitment from pediatric emergency department and urgent care clinics, assessment in a controlled environment.

Participants:

146 pmTBI assessed 7.4±2.3 days and approximately four months post-injury; 104 age/sex matched healthy controls (HC) at equivalent time-points.

Design:

Prospective cohort study.

Main Measures:

Neurosensory exam based on sequence of 10 established tests of vestibular-ocular, oculomotor, vestibulo-spinal and visual functioning.

Results:

The amount of symptom provocation (positive change from pre-test symptomatology) was significantly increased in pmTBI relative to HC on every subtest one week post-injury, as were deficits in monocular accommodative amplitude and King-Devick Test errors. However, symptom provocation didn’t meaningfully alter diagnostic sensitivity/specificity relative to more easily obtained pre-test symptom ratings. Evidence of clinically significant symptom provocation one week post-injury improved sensitivity (Δ=+12.9%) of identifying patients with persistent post-concussive symptoms four months post-injury on an independent symptom measure.

Conclusions:

The diagnostic sensitivity/specificity of neurosensory testing in acutely concussed youth may be limited at one week post-injury as a function of natural recovery occurring in most emergency department cohorts. Neurosensory screening may have greater utility for identifying patients who experience delayed recovery.

Introduction

Public concern is increasing regarding the incidence and the potential long-term effects of pediatric mild traumatic brain injury (pmTBI), including neurosensory (herein defined as vestibular-ocular, oculomotor, auditory, visual, and vestibulo-spinal) sequelae.^1,2 Unlike post-concussive symptoms (PCS), and more similar to cognitive domains, neurosensory abnormalities can be tested using quantifiable metrics (e.g., distance to near point of convergence, response time, test errors) and via self-report.^2–4 Alterations in neurosensory functioning have been shown to have diagnostic (differentiating patients from controls) utility,^5–7 and to be prognostic for academic problems,⁸ cognitive deficits,⁹ delayed return to school¹⁰ and prolonged recovery in athlete samples.^7,11,12

The Vestibular/Ocular Motor Screen (VOMS) is a widely used instrument that examines symptom provocation (i.e., a change in symptoms from pre-test state; see Supplemental Figure 1) within the context of several vestibular and oculomotor challenges.⁵ Previous work in sport-related concussion cohorts suggests that symptom provocation on tests of visual motion sensitivity and horizontal/vestibular-ocular reflex, as well as receded near point of convergence, are the best diagnostic indicators from the VOMS.⁵ In contrast, symptom provocation during smooth pursuit and saccades are reported to be better predictors of medical clearance to return to sports activities, although provocation was not significant when considered conjointly with neurocognitive and symptom measures.¹¹ However, recent work suggests that point-of-care (ED versus concussion clinic) may be an important but understudied variable influencing deficits detected upon neurosensory testing.¹³ Other concussion tools commonly used to assess neurosensory deficits include the King-Devick Test^14,15 and tandem gait, which may have different sensitivities to various aspects of pmTBI.^13,16

To date, few studies have examined the diagnostic sensitivity/specificity of neurosensory abnormalities in children presenting to the emergency department (ED) with pmTBI relative to healthy controls (HC), or determined whether neurosensory abnormalities exhibit prognostic utility beyond demographic variables and symptom self-report.¹⁷ Moreover, the diagnostic sensitivity/specificity (differentiating patients with pmTBI from HC on a single-subject basis) and prognostic (predicting those pmTBI patients who will have poor outcomes) accuracy of individual neurosensory subtests has not been determined for ED samples, a critical requirement for making exams more clinically feasible. Based on previous studies in athletes,^5,18 we predicted that performance on tests of visual motion sensitivity, horizontal/vestibular ocular reflex, and near point of convergence would be roughly equivalent to the entire neurosensory exam in terms of distinguishing pmTBI patients from matched HC recruited from the community. Classifications were performed at the single subject-level in addition to more traditional group statistics to fully delineate clinical utility. Finally, we also predicted that symptom provocation during the sub-acute phase would be associated with recovery (symptom report and quality of life) at four months post-injury, above and beyond a clinically derived risk score¹⁷ and pre-test symptoms.

Materials and Methods

Participants

A total of 147 pmTBI participants (8 – 18 years old) were consecutively recruited from local ED and urgent care settings in this prospective cohort design. Participants were evaluated during the sub-acute (approximately one week post-injury) and early chronic (approximately four months) injury phases that more closely mimicked an out-patient setting. The selected follow-up time corresponds to when most individuals are evaluated for return to play and/or learn. Inclusion criteria were based on American Congress of Rehabilitation Medicine and the Zurich Concussion in Sport Group¹⁹ guidelines. Specifically, all patients with pmTBI experienced head trauma resulting in a Glasgow Coma Score ≥ 13, an alteration in mental status or at least two new symptoms, a loss of consciousness (if present) < 30 minutes, and post-traumatic amnesia (if present) limited to 24 hours. In addition, 106 statistically matched (age/sex) healthy controls (HC) were recruited from the local community using flyers and direct recruitment methods, as well as approaching appropriately aged siblings of patients. The matching was achieved through monthly monitoring of the age and biological sex ratio of pmTBI (male relative to female), followed by targeted recruitment of HC in each age (i.e., within 1 year) and sex category. HC were evaluated at equivalent time points to control for effects related to neurodevelopment and/or repeat assessment.

Exclusion criteria for both patients and HC included history of 1) neurological diagnosis, 2) previous head injury with > 30 minutes loss of consciousness, 3) developmental disorder (autism spectrum disorder or intellectual disability), 4) history of any psychiatric disorders other than adjustment disorder, 5) non-English speaking or 6) history of substance abuse/dependence. HC were also excluded if diagnosed with attention deficit hyperactivity disorder or a learning disability to simulate a typically developing sample, or for a head injury within the past six months. Urine-based drug screens were conducted for all participants at both visits. A positive urine screen was exclusionary for HC, whereas patients with pmTBI (N=4) were not excluded for positive marijuana screens. All results remained fundamentally unchanged when patients with positive screens were excluded from principal analyses. All participants provided informed consent according to institutional guidelines at the University of New Mexico School of Medicine.

One patient with pmTBI and 2 HC were excluded due to completely missing data during the sub-acute visit, resulting in a total sample of 146 patients (67 females; age 14.5±2.5; 7.4±2.3 days post-injury) and 104 HC (47 females; age 14.5±2.5). See Supplemental Table 1 for missing data unique to each subtest. At the time of this investigation, 125/146 patients and 89/104 HC were eligible for their four-month follow-up (i.e., primary study is ongoing). Of these participants, 105 with pmTBI (84% retention; 123.5±13.0 days post-visit 1 date) and 83 HC (93.3% retention; 122.8±11.8 days post-visit 1 date) completed their four-month visit.

Clinical Exam

The current neurosensory exam has been previously described¹³ and focuses on vestibular-ocular, oculomotor, vestibulo-spinal and visual deficits (see Supplemental Materials and Supplemental Figure 1 for full details and test order). Briefly, the exam required participants to first perform a dorsiflexion of both feet for 20 seconds as a non-neurosensory measure of symptom provocation. The majority of the following neurosensory tests (e.g., smooth pursuit, horizontal/vertical saccades, horizontal/vertical vestibular-ocular reflex, visual motion sensitivity) were directly derived from the VOMS⁵ with no alterations to the instructions.

Near point of convergence (distance measured to half centimeter across 3 trials) was acquired with the Astron accommodative rule (Gulden Ophthalmics, Elkins Park, PA) using a standard single 20/30 column as the visual target.²⁰ Each trial ended when participants reported doubling of the stimulus or exhibited loss of binocular convergence. In addition, monocular accommodative amplitude (single trial for each eye reporting distance to blurring of vision) was also measured using the same visual stimulus and accommodation rule. An electronic version of the King-Devick Test (with resultant completion time and number of errors) was also obtained.^14,15 A tandem gait task assessed balance (5 steps forwards and backwards with eyes open and closed), with loss of balance errors recorded when participants stepped away from a taped line.

All participants rated four symptoms (headache, dizziness, nausea, and fogginess) prior to the neurosensory exam (defined as pre-test PCS) and immediately following each sub-test using an 11-point Likert scale.⁵ Although both total and change score methods have been evaluated,¹⁸ symptom provocation was quantified by summing positive differences (i.e., increased symptoms) on each individual test relative to pre-test PCS in the current study. Primary outcome measures were pre-test PCS and symptom provocation, with the latter score excluding the double dorsal foot stretch (Supplemental Figure 1). Seven quantifiable measures examining visual (near point of convergence distance, left and right monocular accommodative amplitude distance, King-Devick errors, King-Devick time) and balance (errors on tandem gait forwards or backwards, eyes open or closed) function were also collected.

Finally, all participants rated concussion symptom severity for retrospective (estimated one month prior to injury), sub-acute and early-chronic time periods using the Post-Concussion Symptom Inventory (PCSI), as well as completed a semi-structured questionnaire about history of previous concussions and a quality of life questionnaire (Pediatric Quality of Life Inventory-PedsQL Generic Core).

Analytic Plan

Patients and HC were compared on key demographics to ensure effective group matching using parametric or non-parametric tests as appropriate. Statistical tests were first conducted to determine whether differences existed between pmTBI and HC at the group level. These group level comparisons were performed with generalized linear models using negative binomial or gamma distributions based on information criterion statistics. Negative binomial distributions accommodate zero heavy counts that lead to over-dispersion relative to Poisson distributions, and more accurately reflect the underlying distribution for both symptom scores and error data.

The second series of analyses used logistic regression and “0.632 bootstrap” classification (B=250 bootstrap resamples) to examine whether certain subtests of the neurosensory exam were more sensitive and/or specific for correctly classifying participants at the single-subject level. These analyses were conducted because significant group differences can exist in larger N studies in the absence of clinically meaningful information at the single-subject level. Bootstrap analyses examined all possible permutations of either symptom provocation following each of the 10 subtests (2¹⁰ = 1024 models) or more quantifiable measures (2⁷ = 128 models) to empirically determine the sensitivity/specificity of a shorter battery. Unlike group-wise comparisons, these analyses were only performed on the subset of individuals completing all aspects of the neurosensory exam. All bootstrap classification rates can be viewed as approximately a repeated 2-fold (i.e., 50–50) cross-validation to ensure generalization of findings.

The last series of analyses used hierarchical models¹³ to prognosticate whether recommended clinical cutoffs for symptom provocation (≥2 on any neurosensory subtest)⁵ during the sub-acute phase were associated with increased overall symptom burden (PCSI; binary logistic) or decreased quality of life (PedsQL; GLM) at four months post-injury. These analyses were therefore limited to pmTBI only. Both models evaluated symptom provocation relative to pre-test symptom and clinical risk scores.¹⁷ Patients with persistent PCS at the early chronic visit were defined based on the 95^th percentile (Z > 1.64) of the log-transformed and distribution corrected standardized PCSI data from HC at the early-chronic visit.

Results

Demographics

No significant differences in age (p = 0.980) or sex (p = 0.913) were observed between groups (see Table 1). A significantly (χ² = 12.14, p < 0.001) greater number of patients (34/146; 23.3%) reported prior concussion history relative to HC (7/104; 6.7%).

Table 1.

Demographics, neurosensory information, and clinical outcome measures.

Demographics	HC (N = 104)	pmTBI (N = 146)	p-value
Age	14.5 (13–16)	14.5 (13–16)	0.980
Sex (% F) ǂ	45.2%	45.9%	0.913
Percent with previous mTBIs	6.7%	23.3%	< 0.001
Days post-injury - SA	-	8 (6–9)
Days post-injury - EC	-	129 (122–139)
Days between visits	120 (114–131)	122 (114–129)	0.745
PPCS-RS	3 (2–4)	5 (3–7)	< 0.001
Neurosensory Symptoms - SA
Pre-test Symptomatology	0 (0–1)	3 (0–9)	< 0.001
Non-specific Symptom Provocation (# Participants) ǂ	2	11	0.001
Total Symptom Provocation	0 (0–1)	2 (0–11)	< 0.001
Objective Measures - SA
Near Point of Convergence (cm)	5.5 (4.17–7)	6 (4.5–8.67)	0.070
Accommodative Amplitude (cm)
Left Eye	8 (6.5–9.5)	8 (6.5–10)	0.111
Right Eye	7.5 (6.5–9)	8 (6.5–10.5)	0.004
Tandem Gait Errors
Forward	1 (0–2)	1 (1–3)	0.079
Backward	2 (1–4)	2 (1–4)	0.692
King-Devick Test
Completion Time	65.85 (54.27–72.78)	66.6 (58.5–78.2)	0.101
Errors	0 (0–0.25)	0 (0–1)	0.003
Clinical Measures- EC	HC (N = 83)	pmTBI (N = 105)
PCSI Total	3 (1–8)	10 (1–26)	< 0.001
PedsQL Total	90.76 (83.7–95.65)	83.7 (75–93.48)	0.001

Note: Median (Interquartile ranges) are presented for all variables unless otherwise specified by the symbol ǂ, in which statistics are specified in field. Non-specific Symptom Provocation is calculated based on increase in symptom ratings on Double Dorsal Foot Stretch. Abbreviations: EC: early chronic; F: female; HC: healthy control; PCSI: Post-Concussion Symptom Inventory; PedsQL: Pediatric Quality of Life Inventory – Generic Core; pmTBI: pediatric mild traumatic brain injury; PPCS-RS: Persistent Post-Concussive Symptoms risk score ranged between 1 and 12 (calculations based on recommendations from Zemek et al., 2016); SA: sub-acute.

Symptom Provocation

At the group level, patients showed significantly higher ratings for both pre-test (Wald = 62.58; p < 0.001; Figure 1A) and total symptom provocation (Wald = 49.15; p < 0.001; Figure 1C). However, several individuals from both cohorts had scores of 0 for either pre-test symptoms (HC: 72/104, 69.2%; pmTBI: 46/146, 31.5%) or symptom provocation scores (HC: 75/104, 72.1%; pmTBI: 62/146, 42.5%), establishing an upper limit on overall classification accuracy. Conversely, the proportion of participants who reported symptom provocation scores ≥ 2 on any single neurosensory test were also limited (HC: 13/104, 12.5%; pmTBI: 58/146, 39.7%). Among patients with pmTBI, days post-injury was significantly inversely related to pre-test symptomatology (r = −0.231; p = 0.005; Figure 1A), but not symptom provocation (rho = −0.106; p = 0.202; Figure 1C). Symptom provocation was significantly greater for patients across all ten neurosensory subtests following Bonferroni correction (p < 0.005; Figure 1D). Both pre-test symptoms (rho = 0.63; p < 0.001) and symptom provocation (rho = 0.41; p < 0.001) were related to an independent measure of overall symptom burden (PCSI) sub-acutely, thus establishing convergent validity.

Figure 1:

In Panel A, box-and-scatter plots display the sum (∑) of pre-test symptomatology in healthy controls (HC; grey, unnotched boxes) and patients with pediatric mild traumatic brain injury (pmTBI; black, notched boxes). The relationship between pre-test symptoms and days-post injury is also presented for patients with pmTBI only. The 95^th percentile for sum of pre-test symptoms in HC is represented by a dashed line. Panel B shows non-specific symptom provocation (Prov; i.e., positive change [Δ]) following the double dorsal foot stretch (DDFS). Panel C depicts total symptom provocation (sum positive change in symptoms from pre-test) and its relationship with days post-injury. Box-and-scatter plots in Panel D depict the sum of positive changes in symptoms for individual neurosensory tests, with red dots indicating data points for individuals who reported symptom provocation following DDFS. Individual subtests are presented in order of administration: SMP = smooth pursuit; SA = saccades; VOR = vestibular-ocular reflex; VMS = visual motion sensitivity; NPC = near point of convergence; MAC = monocular accommodation; TG = tandem gait; KD = King-Devick test.

The difference in symptom provocation during a non-specific test (double dorsal foot stretch) was significant (Wald = 10.30; p = 0.001; Figure 1B), with 7.5% of patients (11/146) and 1.9% of HC (2/104) having non-zero scores. Importantly, within the pmTBI group, non-specific symptom provocation was significantly correlated with total symptom provocation (rho = 0.325, p < 0.001), but not pre-test symptoms (rho = 0.106, p = 0.203). Notably, no participant in either group with a non-zero double dorsal foot stretch score reported zero total symptom provocation (red dots in Figure 1). The two HC with a non-zero double dorsal foot stretch score also exhibited the greatest level of symptom provocation across all sub-tests.

Binary logistic regression indicated that pre-test symptoms (Wald = 25.88; p < 0.001; total accuracy = 68.4%) significantly predicted classification of patients with pmTBI (sensitivity = 86/146; 58.9%) from HC (specificity = 85/104; 81.7%). The addition of total symptom provocation significantly improved model fit (Wald = 6.26; p = 0.012), without meaningfully affecting overall classification accuracy beyond pre-test symptoms (Δ = −0.4%) due to decreased specificity for HC (80/104; Δ = −4.8%) and slightly increased sensitivity for patients with pmTBI (90/146; Δ = +2.7%). Only provocation following the visual motion sensitivity subtest (p = 0.001) significantly improved model fit in addition to pre-test symptoms following correction for multiple comparisons. However, overall classification accuracy was only marginally changed (Δ = 1.3%), with differential performance again noted for individual groups (HC specificity: 81/104; Δ = −3.8%, pmTBI sensitivity: 87/137; Δ = +5.1%). Vertical vestibular-ocular reflex and near point of convergence were significant (see Supplemental Materials) when utilizing rank transformed data¹³ but this did not significantly alter logistic regression results. The top five bootstrap results (Table 2A) from all possible sub-test combinations indicated roughly equivalent classification accuracy as total symptom provocation, with each of the top five models including symptom provocation in response to visual motion sensitivity and vertical vestibular-ocular reflex testing along with one or two other subtests.

Table 2:

Classification Metric Table for Symptom Provocation and Quantifiable Measures

A) Symptom Provocation: Bootstrap Classification
Variables	Overall	Sensitivity	Specificity
Base: Pre-test Symptomatology Only	69.5%	66.1%	73.5%
Top 5 Models: Symptom Provocation and Pre-test
1) VOV+VMS+SMP	71.6%	68.8%	74.9%
2) VOV+VMS+KD+SMP	71.6%	68.3%	75.4%
3) VOV+VMS+KD+SAH	71.5%	68.5%	74.9%
4) VOV+VMS+KD+SAV	71.4%	68.4%	75.0%
5) VOV+VMS+KD+SMP+SAV	71.4%	68.3%	75.1%
B) Quantifiable Measures: Bootstrap Classification
Variables	Overall	Sensitivity	Specificity
Top 5 Models: Quantifiable Metrics and Pre-test
1) NPC+MAC-L+MAC-R+KD-T+KD-E+TGB-E	70.5%	67.3%	74.8%
2) NPC+MAC-L+KD-T+TGF-E+TGB-E	70.3%	67.3%	74.3%
3) NPC+MAC-L+MAC-R+KD-E+TGB-E	70.3%	67.1%	74.7%
4) NPC+MAC-L+MAC-R+KD-T+TGB-E	70.2%	67.0%	74.4%
5) MAC-L+MAC-R+KD-T+KD-E+TGB-E	70.1%	66.9%	74.4%

Note that bootstrapping results were primarily used to determine the utility of a shorter battery. They are therefore based on data from participants who completed all aspects of the neurosensory exam (i.e., no missing data). Abbreviations— KD-T/KD-E: King-Devick Test (Completion time/Errors); MAC-L/MAC-R: Monocular Accommodative Amplitude (Left Eye/Right Eye); NPC: Near point of convergence; SAH/SAV: Saccades (Horizontal/Vertical); SMP: Smooth pursuit; TGF-E/TGB-E: Tandem gait errors (Forward/Backward); VMS: Visual motion sensitivity; VOV: Vestibular-ocular reflex (Vertical).

Quantifiable Metrics

A series of seven generalized linear models examined group differences on quantifiable neurosensory measures (Figure 2). Only right monocular accommodative amplitude (Wald = 8.34, p = 0.004) and King-Devick Test errors (Wald = 9.05, p = 0.003) demonstrated significant group effects (pmTBI > HC) following Bonferroni correction (p < 0.007). Hierarchical binary logistic regression indicated similar pre-test symptom results (Wald = 25.85; p < 0.001) for total accuracy (68.9%), pmTBI sensitivity (83/139; 59.7%) and HC specificity (83/102; 81.4%). Quantifiable measures did not improve classification (all p’s > 0.10; total accuracy Δ = −0.4%) above pre-test symptoms when all seven were entered simultaneously or individually. Bootstrapped classification accuracy indicated roughly equivalent classification accuracy for all quantifiable tests entered simultaneously versus the top five models. All top five models (Table 2B) included left monocular accommodative amplitude and backward tandem gait errors, with near point of convergence, right monocular accommodative amplitude, and King-Devick response time and error metrics present in other models.

Figure 2:

Box-and-scatter plots display the quantifiable raw data for the neurosensory tasks across groups (HC = healthy control; pmTBI = patients with pediatric mild traumatic brain injury). Plots depict: average distance in centimeters (cm) for the near point of convergence distance (NPC; Panel A); King-Devick (KD) completion time (Panel B) in seconds (s); KD error count (Panel C); monocular accommodation amplitude (MAC) for left and right eyes (Panel D); and tandem gait (TG) error count (Panel E) for forward and backward walking. Asterisks denote significance following Bonferroni correction.

Prognostic Accuracy

The last analyses examined whether symptom provocation during the sub-acute phase was associated with long-term outcomes for symptom burden or quality of life in patients with pmTBI only. Results indicated that pre-test symptoms accounted for significant unique variance in the base model (Wald = 9.13, p = 0.003; overall accuracy = 74.8%), with higher specificity for recovered patients (69/72; 95.8%) relative to sensitivity for those with persistent PCS (8/31; 25.8%). The addition of the clinically relevant symptom provocation score at one week post-injury significantly improved model fit (Wald = 5.69, p = 0.017) and increased overall classification accuracy at four months (Δ = +3.8%). Importantly, this resulted from improved sensitivity for classification of persistent PCS patients (12/31; Δ = +12.9%) rather than specificity to recovered patients (Δ = 0.0%). Analysis focused on quality of life similarly indicated that pre-test symptoms accounted for significant unique variance (Wald = 9.65, p = 0.002), whereas symptom provocation did not improve model fit (p > 0.10) or meaningfully alter classification accuracy.

Discussion

The primary aims of this study were to examine diagnostic (patients versus controls) and prognostic (recovered versus non-recovered patients) accuracy of a neurosensory screening exam in a cohort of recently concussed patients with pmTBI. To maximize the clinical translation of results, our statistical plan investigated both group differences (i.e., group central tendency) and classification accuracy (i.e., at single-subject level) for symptom provocation relative to more easily obtained pre-test symptoms and clinically derived risk scores.¹⁷ Results indicated significantly greater symptom provocation in patients compared to HC for every subtest of the VOMS, as well as group differences on measures of monocular accommodative amplitude and King-Devick Test errors. In spite of these group differences, only symptom provocation following visual motion sensitivity and vestibular-ocular reflex subtests improved model fit beyond pre-test symptoms at one week post-injury. Neither test nor the total symptom provocation score improved the ability to differentiate patients from HC at a single-subject level (diagnostic accuracy) in this ED sample. However, clinically significant symptom provocation⁵ at one week post-injury did improve prognostic accuracy for patients with persistent PCS at four months post-injury.

The null finding for total symptom provocation on diagnostic classification accuracy could be the result of several factors. Specifically, our previous study¹³ demonstrated modest improvement in diagnostic classification accuracy relative to pre-test symptoms in patients with pmTBI identified in an ED setting, with more significant change among patients identified in a subspecialty concussion clinic. Previous studies on neurosensory abnormalities have predominantly focused on sport-related concussion^5,18 or subspecialty concussion clinic^3,6 cohorts and emphasized group-level differences rather than classification accuracy, two very different statistical concepts. Second, pmTBI is known for its heterogeneous clinical presentation,^11,21 and simply summing provocation across all subtests may not accurately capture the diverse aspects of symptom provocation. Finally, 31.5% of our pmTBI sample did not report pre-test symptoms at approximately one week post-injury, and only 39.7% of patients demonstrated clinically significant symptom provocation (2 or greater on any subtest).⁵ Conversely, similar to previous results,²² 30.8% of HC endorsed some pre-test symptoms due to their non-specific nature. Importantly, these proportions statistically limit the level of achievable diagnostic accuracy despite significant group differences on every subtest. Diagnostic sensitivity and specificity for pmTBI should typically decline as a function of time post-injury²³ due to the natural recovery process.

The goal of most screening exams is to obtain clinically useful information as quickly as possible.^5,7,11,12 Bootstrap results indicated that diagnostic accuracy for symptom provocation following the visual motion sensitivity subtest, vestibular-ocular reflex subtest, and monocular accommodative amplitude was similar to the entire screening battery. Symptom provocation following visual motion sensitivity also statistically improved diagnostic accuracy relative to pre-test symptoms, with similar results for vestibular-ocular reflex depending on the statistical transformation of the data. These results partially replicate previous findings from sport-related concussion samples and highlight the utility of these two subtests for diagnostic classification⁵ and their potential to predict longer recovery times.¹⁸ In contrast, the visual motion sensitivity and vestibular-ocular reflex subtests also had the lowest completion rates due to cervical injury comorbidities and movement restrictions (e.g., use of Miami J collars). From the quantifiable metrics, tandem gait errors, monocular accommodative amplitude, near point of convergence, and errors on the King-Devick Test produced the best classification accuracy during bootstrap testing, with monocular accommodative amplitude and King-Devick errors also demonstrating significant group differences approximately one week post-injury. If these findings are replicated, clinicians could consider using a subset of these neurosensory tests to shorten administration time in clinical settings.

Previous work suggests that a minority of youth and collegiate-level healthy athletes exhibit single-test symptom provocation scores of ≥ 2 during VOMS administration.^5,18,24 Results from our community sample of HC were similar (12.5% with symptom provocation ≥ 2), although still approximately double commonly recommended false positive rates (i.e., 5%). Differential testing of effort and symptom specificity has important clinical ramifications in cognitive testing following pmTBI²⁵ but is not routinely utilized in other neurosensory screens. A non-neurosensory measure of symptom provocation (double dorsal foot stretch) is unique to this exam¹³ and was significantly elevated in patients with pmTBI relative to HC. Non-specific symptom provocation further correlated with neurosensory symptom provocation following pmTBI. Other studies have reported increased provocation during physical exertion²⁶ and cognitive testing,²⁷ suggesting that provocation may be non-specific in nature for certain individuals rather than indicative of exacerbation during a specific stressor (neurosensory or otherwise).

A primary challenge for clinicians is to identify children who will develop persistent PCS following pmTBI.^17,28 A recently developed clinical risk score (9 demographic and symptom scores variables) has approximately 70% accuracy for predicting prolonged recovery at one month post-injury in ED and concussion clinic settings.^17,29 Previous studies have similarly associated neurosensory test results with recovery,^7,11,30,31 although predictive values decreased when considered jointly with commonly utilized neurocognitive and symptom measures.^7,11 Our results revealed a small but statistically significant improvement in classification accuracy for symptom provocation scores obtained at one week post-injury relative to both pre-test symptoms and clinically derived risk scores, with classification improvement primarily occurring for patients with persistent PCS. Importantly, the concept of “recovery” has also been shown to vary across neurosensory and cognitive domains,^11,21 indicating the need for a cautious approach when classifying patients into binary categories for a disorder that is notorious for its heterogeneous clinical presentation.^5,12,18

Several limitations of the current study should be noted. Foremost, the sample size was too small to detect smaller effects or examine relationships between non-specific symptom exacerbation and long-term prognosis. Second, derivation of Zemek’s 5P clinical risk score was based on more acute assessments than utilized in the current study.¹⁷ However, subsequent studies have validated the 5P criteria in one week post-injury assessments performed in concussion clinic samples.²⁹ Third, test administrators were not blinded to each participant’s diagnosis. Finally, this study utilized change scores based on previously published recommendations¹⁸ and high correlation between change and total symptom score.¹³ However, clinical results may vary as a function of how symptom provocation is calculated and subsequently modeled. This is critical given the sparse nature of symptom provocation in HC in combination with the typical recovery from symptoms that occur for most patients more than one month post-injury.

In summary, neurosensory, cognitive, and physical challenges replicate a child’s real-world experiences (e.g., school and active play) where symptoms may be exacerbated during periods of increased activity, stimulation and stress.^5,13 Symptom provocation may prove useful for assessing the extent of recovery from pmTBI, but requires replication across multiple points of care and methods of statistical evaluation. Importantly, the clinical translatability of neurosensory tests will ultimately depend on ease of administration, the capacity to provide a quantifiable metric versus self-report, and the ability to predict outcomes or persistent PCS sub-types above and beyond more simply obtained measures. To this end, vertical/horizontal saccades, smooth pursuits, and tandem gait forward subtests did not account for substantial changes in diagnostic or prognostic classification accuracy. These subtests may therefore not be essential aspects for a briefer neurosensory exam in ED patients examined one week post-injury. In contrast, visual motion sensitivity and vestibular-ocular reflex appear to show the most promise for developing shorter batteries that provide similar classification accuracy, with these findings now replicated across multiple independent samples and points of care.^5,18