Optimizing the Combination of Common Clinical Concussion Batteries to Predict Persistent Postconcussion Symptoms in a Prospective Cohort of Concussed Youth

Abstract

Background:

Studies have evaluated individual factors associated with persistent postconcussion symptoms (PPCS) in youth concussion, but no study has combined individual elements of common concussion batteries with patient characteristics, comorbidities, and visio-vestibular deficits in assessing an optimal model to predict PPCS.

Purpose:

To determine the combination of elements from 4 commonly used clinical concussion batteries and known patient characteristics and comorbid risk factors that maximize the ability to predict PPCS.

Study Design:

Cohort study; Level of evidence, 2.

Methods:

We enrolled 198 concussed participants–87 developed PPCS and 111 did not–aged 8 to 19 years assessed within 14 days of injury from a suburban high school and the concussion program of a tertiary care academic medical center. We defined PPCS as a Post-Concussion Symptom Inventory (PCSI) score at 28 days from injury of ≥3 points compared with the preinjury PCSI score–scaled for younger children. Predictors included the individual elements of the visio-vestibular examination (VVE), Sport Concussion Assessment Tool, 5th Edition (SCAT-5), King-Devick test, and PCSI, in addition to age, sex, concussion history, and migraine headache history. The individual elements of these tests were grouped into interpretable factors using sparse principal component analysis. The 12 resultant factors were combined into a logistic regression and ranked by frequency of inclusion into the combined optimal model, whose predictive performance was compared with the VVE, initial PCSI, and the current existing predictive model (the Predicting and Prevention Postconcussive Problems in Pediatrics (5P) prediction rule) using the area under the receiver operating characteristic curve (AUC).

Results:

A cluster of 2 factors (SCAT-5/PCSI symptoms and VVE near point of convergence/accommodation) emerged. A model fit with these factors had an AUC of 0.805 (95% CI, 0.661-0.929). This was a higher AUC point estimate, with overlapping 95% CIs, compared with the PCSI (AUC, 0.773 [95% CI, 0.617-0.912]), VVE (AUC, 0.736 [95% CI, 0.569-0.878]), and 5P Prediction Rule (AUC, 0.728 [95% CI, 0.554-0.870]).

Conclusion:

Among commonly used clinical assessments for youth concussion, a combination of symptom burden and the vision component of the VVE has the potential to augment predictive power for PPCS over either current risk models or individual batteries.

Concussion is a common injury in children and adolescents.² One of the biggest challenges for those who care for concussed youth is identifying those at higher risk for prolonged symptoms and, thus, in need of more targeted intervention.²⁵ Although the majority of concussed youth will achieve symptom resolution within 1 month of injury, approximately 30% will develop persistent postconcussion symptoms (PPCS), defined as symptoms above the baseline persisting >28 days.^17,46 The vast majority of youth who show symptoms at 1 month continue to be symptomatic 3 months after injury,¹² with significant physical, cognitive, and mental health sequelae.^10,35,43 Recent studies have shown that implementation of concussion-specific therapies—such as aerobic exercise protocols and visio-vestibular rehabilitation—can help expedite recovery when prescribed early in the course of injury,^20,22,23 making acute identification of those at the highest risk for PPCS of critical importance.

Previous studies have identified individual risk factors associated with prolonged symptoms. These include age, with children and adolescents showing longer recovery than adults^13,40 and older adolescents showing longer recovery than younger children⁴⁶; sex, with female youth showing prolonged recovery time when compared with male youth^9,31; and certain comorbidities, such as previous concussion history^8,31 and a history of migraine headaches,³² associating with longer recovery. In addition, elements of commonly used clinical concussion batteries, such as symptom scales,³⁰ and balance testing, such as the modified Balance Error Scoring System (mBESS)⁴⁶ and visio-vestibular testing deficits,^7,27 have all individually shown associations with prolonged symptom recovery in concussed youth.

The largest study to combine these risk factors into a single predictive model in the youth population to date is the Predict and Preventing Postconcussive Problems in Pediatrics (5P) prediction rule, which focuses primarily on symptom scores and patient characteristics/comorbid features.⁴⁶ Although representing a significant improvement over physician judgment alone, the 5P prediction rule retains only moderate discrimination for PPCS—an area under the receiver operating characteristic curve (AUC) of 0.68 in the validation cohort. This discriminatory metric has been replicated in subsequent studies.^17,46 To improve our predictive ability, we hypothesized that the addition of visio-vestibular deficits to known patient characteristics, comorbid risk factors, and individual elements of commonly used concussion batteries—such as the Sports Concussion Assessment Tool, 5th edition (SCAT-5) or the King-Devick (K-D) test—would result in a more effective method for stratifying risk for PPCS in concussed youth.

Therefore, this study aimed to determine the optimal combination of elements from 4 commonly used clinical concussion batteries—(1) the visio-vestibular examination (VVE); (2) the SCAT-5; (3) the K-D test; and (4) a validated symptom scale, the Post-Concussion Symptom Inventory (PCSI)—with known patient characteristics and comorbid features to maximize the ability to stratify risk for concussed youth for PPCS using analytic techniques to adjust for the significant potential overlap of these individual features and demonstrate a method to reduce the overlapping tests currently used to stratify risk.⁴⁷ Secondary objectives included comparing the predictive power of the optimal model developed in this analysis to both the 5P prediction rule and the individual batteries in isolation.

METHODS

Study Design and Cohort

We recruited participants aged 8 to 19 years between August 2017 and June 2022 as part of a large prospective, observational study of both concussed and nonconcussed youth, as described in previous work.⁵ Although developmental differences do exist, children were included as young as 8 years, given the understudied nature of the 8- to 12-year-old population,²⁶ as well as their inclusion of the 5P rule.⁴⁶ We recruited participants as a convenience sample both from a local suburban middle and high school and the multidisciplinary concussion program of our tertiary care children’s hospital. In both settings, the diagnosis of concussion was made by the treating provider, a sports medicine pediatrician (C.L.M.), in accordance with the most recent International Consensus Statement on Concussion in Sport.²⁸ At the first visit, as part of the large observational study, multiple clinical assessments were performed, including those listed in the following sections. Participants were observed at each subsequent clinical visit through clearance, as determined by the sports medicine pediatrician. To be included in the study, participants had to have their first clinical visit within 14 days of injury and have ≥2 visits during their clinical course. Participants with only 1 visit in the first 31 days after injury and those missing either a preinjury symptom score or a symptom score at the first assessment were excluded from the analysis. In addition, participants with unknown PPCS values (see Outcome Assessment section) were excluded from the primary analysis. Participants who were in active recovery from a previous concussion (≤30 days of clearance) or those who had lower extremity trauma that would preclude gait or balance assessment were also excluded. Before enrollment, participants aged ≥18 years or guardians (for participants aged <18 years) provided written consent, with those aged <18 years providing verbal assent. This study was approved by our institution’s institutional review board.

Clinical Assessments

All clinical assessments were obtained and performed by trained research staff at each clinical visit. Only the clinical assessments at the first visit were utilized for our predictor elements.

Patient Characteristics/Clinical Covariates.

We collected age, sex, race, and ethnicity from the electronic medical record abstraction as well as previous concussion history and diagnosed history of migraine headache from a patient self-report form completed by the patients or their parent/guardian. We included concussion history and migraine headache history based on their significance in the 5P study.⁴⁶ Symptom duration was recorded as a continuous variable. However, in line with the 5P rule, we dichotomized this variable to either ≥7 days or <7 days.⁴⁶

Visio-Vestibular Examination.

The VVE is a clinical assessment of vision and vestibular function and includes the following examination elements⁴: smooth pursuit; horizontal and vertical saccades; horizontal and vertical gaze stability, or the angular vestibulo-ocular reflex (VOR); the near point of convergence (NPC); right and left monocular accommodative amplitude; and complex tandem gait. See Table 1 for a description of each examination element and its abnormalities. In addition, a video demonstration of the VVE is available at https://www.youtube.com/watch?v=lGUDZnZOieM

TABLE 1.

Description of the 9 Individual VVE Elements^a

Individual Examination Element	How to Perform	Abnormalities
[1] Smooth pursuit	Examiner moves the finger in the horizontal plane, increasing speed, for 5 repetitions, evaluating for abnormal signs and symptom provocation	Abnormal signs: Jerky/jumping eye movements, >1 beat of nystagmus Symptom provocation: Headache, nausea, dizziness, eye fatigue, eye pain 3
[2] Horizontal and [3] vertical saccades	The examiner holds 2 fixed objects (the fingers) approximately shoulder width (horizontal) and forehead-to-sternal notch distance (vertical) apart and asks the participant to rapidly move his or her eyes between the objects	Symptom provocation (headache, nausea, dizziness, eye fatigue, and eye pain) with ≤20 repetitions41
[4] Horizontal and [5] vertical gaze stability (VOR)	The examiner holds the object (the thumb) fixed and asks the participant to nod his or her head no (horizontal) or yes (vertical) while keeping eyes fixed on the object	Symptom provocation (headache, nausea, dizziness, eye fatigue, and eye pain) with ≤20 repetitions41
[6] Near point of convergence	The examiner uses a standard accommodative rule (Gulden Ophthalmics), via which the participant identifies the distance at which letters on a standard 20/ 30 card break (double vision occurs)	Break occurring at >6 cm38
[7] Left and [8] right monocular accommodative amplitude	The examiner uses a standard accommodative rule (Gulden Ophthalmics), via which the participant, with 1 eye open, identifies the distance at which letters on a standard 20/30 card blur	Abnormal distance is defined by normative values by age using the Hofstetter formula (eg, <9.1 cm is normal at age 8 years, <9.3 cm is normal at age 9, etc)33
[9] Complex tandem gait	Participant walks with 1 foot directly in front of (or behind) the other, for 5 steps each, for 4 conditions: forward eyes open, forward eyes closed, backward eyes open, and backward eyes closed, while the examiner evaluates for steps off the straight line (up to 5) or the presence of sway (raising of arms for stability or any truncal movement)	Total scores 0-24, with 1 point for each step off the straight line and 1 point for the presence of sway for each of the 4 conditions (each condition has a score of 0 to 6) Abnormal is defined as a score of ≥5 out of 246

^aVOR, vestibulo-ocular reflex; VVE, visio-vestibular examination.

Sport Concussion Assessment Tool, 5th Edition.

The SCAT-5 evaluates concentration, memory, attention, symptom burden, and balance. The SCAT-5 is made up of several individual elements. Symptom evaluation includes total symptoms (scaled 0-22) and symptom severity score (scaled 0-132), each rated on a 7-point Likert scale from 0 to 6. Memory—including immediate (scaled 0-15) and delayed (scaled 0-5) word memory—involves asking the participant to recall a list of 5 words across 3 trials for immediate memory. Orientation rated, from 0 to 5, and concentration, rated from 0 to 5, are assessed by asking the participant the year, month, date, day of week, and time of day and then asking the participant to repeat a series of digits in reverse order and list the months of the year in reverse order, respectively. Finally, the mBESS is composed of the double-leg, single-leg, and tandem stances, whereby the participant is asked to stand on a firm surface for each condition, with hands on the hips and eyes closed for 20 seconds, with up to 10 errors for each condition (>10 errors are classified as 10).^5,39 For ease of interpretability, we administered the adolescent SCAT-5 to all participants in the study.

K-D Test.

The K-D test assesses the total time for a participant to read a series of numbers across 3 test cards with increasing difficulty, relating to the offset orientation of the numbers on the card, and evaluates eye-tracking abnormalities.¹⁴

Post-Concussion Symptom Inventory.

We utilized 2 versions—child self-report and adolescent self-report—of the PCSI, a validated symptom report tool. The PCSI adolescent self-report, administered to participants aged ≥13 years, includes 21 concussion-like symptoms. Similar to the symptom scale of the SCAT-5, each symptom is rated on a 7-point Likert scale (from 0 = no symptom to 6 = most severe symptoms). The PCSI child self-report form is administered to children aged <13 years and consists of 17 concussion-like symptoms, and is rated on a 3-point Likert scale (from 0 = no symptoms to 2 = most severe symptoms).³⁷ Four symptom clusters are generated from the PCSI—physical, cognitive, emotional, and fatigue. To standardize the child and adolescent PCSI scores, each child score was proportionally scaled up to match the range of the adolescent scores, an approach utilized in previous studies.¹⁶ See Figure 2 for details of the adjusted scales.

Figure 2.

Factors created by sparse principal component analysis. Twelve factors were made from the 24 elements of the clinical assessments listed in Table 2. The loading of each component is reported in parentheses. The K-D completion time is scored in seconds; SCAT-5 immediate memory is a 0 to 15 scale; SCAT-5 concentration, orientation, and delayed recall are scored from 0 to 5; The SCAT-5 mBESS stances are scored on a 0 to 10 scale; all VVE elements are scored as normal/abnormal; the SCAT-5 symptom score is a 0 to 22 scale; the SCAT-5 symptom severity score is a 0 to 132 scale; the PCSI physical score is a 0 to 48 scale (PCSI child physical score is proportionally scaled from original 0-16 to match the adolescent 0-48 scale); the PCSI fatigue score is a 0 to 18 scale (the PCSI child fatigue score is proportionally scaled from original 0-4 scale to match the adolescent 0-18 scale); the PCSI emotional score is a 0 to 24 scale (the PCSI child emotional score is proportionally scaled from the original 0-6 scale to match the adolescent 0-24 scale); the PCSI cognitive score is a 0 to 36 scale (the PCSI child cognitive score is proportionally scaled from the original 0-8 scale to match the adolescent 0-36 scale). K-D, King-Devick test; mBESS, modified Balance Error Scoring System; NPC, near point of convergence; PCSI, Post-Concussion Symptom Inventory; SCAT-5, Sport Concussion Assessment Tool, 5th Edition; VOR, vestibulo-ocular reflex; VVE, visio-vestibular examination.

5P Prediction Rule.

The 5P prediction rule for PPCS includes the following 9 elements: (1) age—higher risk for participants age ≥13 years versus 8 to 12 years versus <8 years; (2) sex—higher risk for females; (3) previous concussion with symptom duration for ≥1 week; (4) physician-diagnosed migraine history; (5) mBESS tandem stance—≥4 errors—and the symptoms from the PCSI, with a score of ≥1 on the Likert scale equating to a positive test; (6) answering questions slowly; (7) headache; (8) sensitivity to noise; and (9) fatigue. Of note, 3 features in the original model (adolescent age, female sex, and fatigue) were weighted more heavily than the remaining individual features. We included this weighting in our analysis of the original 5P model. However, we also implemented an unweighted version of the 5P rule to include weighting for time since injury, which allows for a more direct comparison with the other time-weighted models in our analysis, given the variability of time from injury present.

Outcome Assessment

The primary outcome of this study was the presence of PPCS. Previous studies have defined PPCS as concussion symptoms persisting beyond 28 days from injury,⁴⁶ with a recommendation to use the preinjury symptom score as the comparison.¹⁵ As most patients had no previous evaluations at our clinic, they completed the preinjury symptom scale at the first visit after injury, in line with previous studies relying on a preinjury symptom scale.⁴⁶ Participants were defined as meeting the criteria for PPCS if their PCSI total score at 28 days from injury was ≥3 points of their reported preinjury score. As data were obtained from clinical visits, not all participants were evaluated exactly 28 days from injury. Therefore, observed values of assessments 28 ± 3 days, with the closest to the 28th day, were used to determine the PPCS status. If the participant was not assessed in the 28 ± 3-day window but was assessed >31 days after injury and had ≥3 points of his or her preinjury score, the participant was defined as having PPCS (Figure 1). If the participant’s last assessment was <25 days after injury and had <3 points of his or her preinjury PSCI score, the participant was defined as not having PPCS. Participants whose last assessment was <25 days after injury and had ≥3 points of their reported preinjury score, and participants who were assessed >31 days and had <3 points of their reported preinjury score, were classified as having the “PPCS unknown” status and were excluded from the primary analysis. To assess the effect of excluding these participants, a sensitivity analysis was conducted in which the PPCS status for these participants was imputed.

Figure 1.

Flow diagram of inclusion and persistent postconcussion symptom (PPCS) status.

Modeling and Statistical Analyses

Patient characteristics of the study cohort were summarized using standard descriptive statistics. The PPCS status was assigned for observed values using the aforementioned criteria. See Appendix 1 (available online) for a detailed discussion of our imputation and modeling approaches. In brief, missing values were observed in the clinical assessment elements and the clinical/patient covariates. Analysis results were averaged over multiple imputations of missing values based on multivariate imputation by chained equations.^36,44 To account for the correlation between the 24 elements of the 4 clinical assessments, we conducted a sparse principal component analysis to group the elements into interpretable factors. To increase interpretability, the number of factors and sparsity level were selected to avoid overlap, ensuring each element contributed to only 1 factor. This resulted in 12 factors that were linear combinations of the 24 elements. Table 2 and Figure 2 show the contributing elements and the combinations (loadings) for each factor.

TABLE 2.

The 24 Elements of the 4 Clinical Assessments^a

VVE	SCAT-5
Smooth pursuit	Total symptom score
Horizontal saccades	Symptom severity score
Vertical saccades	Orientation
Horizontal gaze stability	Immediate memory
Vertical gaze stability	Delayed recall
Near point of convergence	Concentration
Right monocular accommodation	mBESS: double-leg stance
Left monocular accommodation	mBESS: single-leg stance
Complex tandem gait	mBESS: tandem stance
PCSI	K-D test
Somatic score	Total completion time
Vestibular score
Fatigue score
Emotional score
Cognitive score

^aK-D, King-Devick; mBESS, modified Balance Error Scoring System; PCSI, Post-Concussion Symptom Inventory; SCAT 5, Sport Concussion Assessment Tool, 5th Edition; VVE, visio-vestibular examination.

We used a ridge penalized logistic regression model to predict the PPCS status using the 12 factors and 4 additional clinical/patient covariates—age, sex, presence or absence of concussion history, and presence or absence of migraine history. We adapted the model to account for the variability in time from presentation to first visit, t, by using kernel weighting. We implemented forward selection with the AUC as the criterion to determine the most highly predictive features of PPCS. Factors selected as a top factor in ≥1 of the 10 imputed data sets were selected as overall top factors. We then compared the predictive performance of these overall top factors to the VVE and PCSI batteries in isolation and the elements of the 5P prediction rule. Four separate, penalized, weighted logistic regression models predicting PPCS were fit—1 with all elements of VVE, 1 with all elements of PCSI, 1 with all elements of the 5P prediction rule (without element weighting), and 1 with the top factors—and determined by the forward selection procedure. The AUC for each model was calculated as previously described, and confidence intervals for the AUC were calculated based on 1000 bootstrap samples of the training data. In addition, we implemented the original 5P prediction rule, using its weighted factors, and calculated the AUC using the 13 threshold levels of the 0 to 12 risk point scale. All analyses were performed using R Statistical Software Version 4.2.1 (R Foundation for Statistical Computing).

RESULTS

The study cohort included 198 total participants—87 (44%) with PPCS and 111 (56%) without PPCS. Table 3 provides a summary of the patient and clinical characteristics of the study cohort. Overall, 12.9% missingness was observed in the covariates of the 198 participants used in the analysis; these missing values were handled by multivariate imputation via chained equations as previously described.

TABLE 3.

Patient and Clinical Characteristics of the Study Cohort^a

Patient Characteristics	No PPCS (n = 111)	PPCS (n = 87)	P b
Mean age, y	15.4 (1.7)	15 (2.1)	.115
Female sex	65 (58.6)	31 (35.6)	.002
Race/ethnicity			.835
Non-Hispanic White	88 (79.3)	65 (74.7)
Non-Hispanic Black	8 (7.2)	6 (6.9)
Hispanic	4 (3.6)	4 (4.6)
Other	11 (9.9)	12 (13.8)
History of previous concussion	51 (45.9)	46 (52.9)	.410
History of migraines	17 (15.3)	17 (19.5)	.553
Mean days from injury to first visit	6.2 (3.7)	8.6 (3.9)	<.001
Mean PCSI at initial assessment	31 (24.5)	55.8 (24.5)	<.001
Mean preinjury PCSI	12.0 (16.2)	8.7 (10.1)	.089

^aData are presented as mean (SD) or n (%). PCSI, Post-Concussion Symptom Inventory; PPCS, persistent postconcussion symptoms (no return to baseline at 28 days).

^bt tests were used to compare means. Chi-square tests were used to compare proportions. The Fisher exact test was used to compare proportions for race/ethnicity.

Figure 3 displays the results of the forward selection procedure averaged over the 10 imputed data sets. Two factors were selected as overall top factors: SCAT-5 1 PCSI symptoms and NPC 1 accommodation. The SCAT-5 and PCSI symptoms factor was selected in 69.9% of the models with a mean rank of 1.1 and was selected as a top factor in all 10 imputed data sets. The NPC 1 accommodation factor was selected in 60.4% of the models with a mean rank of 2.2 and was selected as a top factor in 9 out of 10 imputed data sets.

Figure 3.

Results of the forward selection procedure with 12 factors and 4 patient/clinical covariates. Scatter plot of the mean number of times a factor is selected into the model out of 100 (y-axis) versus a mean rank in selection order when present in a model (x-axis). Values are averaged across results from 10 imputed data sets. The top factors are highlighted in lighter gray. conc, concussion; hx, history; K-D, King-Devick test; mBESS, modified Balance Error Scoring System; NPC, near point of convergence; PCSI, Post-Concussion Symptom Inventory; SCAT-5, Sport Concussion Assessment Tool, 5th Edition; VOR, vestibulo-ocular reflex.

The sequence of estimated regression coefficients for the SCAT-5/PCSI symptoms factor was positive across the range of t (mean, 0.142), suggesting a positive association between the number of symptoms and PPCS. Similarly, the NPC 1 accommodation factor coefficients were positive across the range of t (mean, 0.118), indicating a positive association between the NPC 1 accommodation abnormalities and PPCS.

Figure 4 shows the predictive performance of the model fit with the top 2 factors compared with models fit with all elements of the PCSI, all elements of the VVE, and all elements of the 5P prediction rule. The model fit with the top factors performed as well as each individual battery and the 5P prediction rule in determining PPCS. It had a higher point estimate of the AUC than any individual batteries or 5P prediction rule; however, the 95% CIs overlapped. The model with the top factors had an AUC of 0.805 (95% CI, 0.661-0.929), compared with the PCSI (AUC, 0.773 [95% CI, 0.617-0.912]), VVE (AUC, 0.736 [95% CI, 0.569-0.878]), and time-weighted 5P prediction rule (AUC, 0.728 [95% CI, 0.554-0.870]). The original 5P prediction rule had an AUC of 0.608 (95% CI, 0.531-0.682).

Figure 4.

AUC of models fit with the top 2 factors from the forward selection procedure, elements of VVE, elements of PCSI, and elements of the 5P prediction rule. Factors refer to the combination of the factor 1 (SCAT-5 1 PCSI symptoms) and factor 2 (NPC 1 accommodation). Error bars represent 95% CIs for the AUC calculated from 1000 bootstrap samples. 5P, Predicting and Preventing Postconcussive Problems in Pediatrics; AUC, area under the receiver operating characteristic curve; NPC, near point of convergence; PCSI, Post-Concussion Symptom Inventory; SCAT-5, Sport Concussion Assessment Tool, 5th Edition; VVE, visio-vestibular examination.

Finally, our sensitivity analysis, which includes an imputation of 48 participants for whom the PPCS status was classified as “unknown,” is shown in Appendix 2 (available online). Overall, including these participants with their imputed values did not change the top 2 factors selected, nor did it significantly change the AUCs of the top factor model or the individual batteries.

DISCUSSION

This study aimed to discern the optimal combination of individual elements of 4 commonly used clinical concussion batteries, in combination with patient and comorbid factors, that most strongly predict PPCS among a group of concussed youth assessed within 14 days of injury. The combination of symptom scales and the vision testing elements (NPC and monocular accommodation) of the VVE provided the largest and most unique contributions to the final model, which, although resulting in a higher point estimate AUC than any individual battery alone, presented overlapping 95% CIs with the model composed of the overall symptom burden.

Symptoms, the current mainstay of diagnosis in youth concussion,²⁸ were found in the present study to be the strongest PPCS predictor, with the symptom factor selected in over three-fourths of models, with a mean rank of 1.1. We must consider both the time frame of symptoms and the population chosen for the present study in this context. Participants were seen a mean of 1 week from injury, and as our population was primarily recruited from our sports medicine specialty clinic, they may have represented a group with a more severe concussion subtype. Given the outcome definition of interest, it is likely that the later a child was evaluated, the more accurately the symptom burden would predict PPCS—that is, symptom burden at day 13 is likely more predictive than at day 2. As discussed later, this has implications for our generalizability, specifically with regard to applying our findings to a more acutely injured population. Several studies over the past decade have shown symptoms to be a strong predictor of prolonged symptoms. For example, in a 2014 study of 531 children and young adults, Meehan et al³⁰ found symptom burden at the initial visit, when compared with age, sex, concussion history, and headache history, to be the strongest predictor of prolonged recovery. In the original 5P prediction rule, 4 of the 9 individual risk factors relate to initial symptom burden, and some of the strongest effect sizes from that study were seen for those 4 individual symptoms.⁴⁶

Symptom burden is a relatively quick and inexpensive assessment that can be completed without significant training, making it an attractive tool to stratify risk; however, previous studies have demonstrated the limitations of symptom assessment, including their lack of specificity for concussion and the subjective nature of symptoms.¹⁸ Given the variability in baseline symptoms of adolescents, many have advocated for a definition of both PPCS and recovery to reference a concussed youth’s preinjury baseline symptom score,¹⁵ as was done in our study. However, obtaining a postinjury self-report of preinjury “baseline” (as few youths will have an actual preinjury baseline symptom score completed) carries with it its limitations of recall bias.⁴² Thus, it is appealing to consider including additional testing elements as part of risk stratification, such as the VVE.

Both the unique and large contributions of the vision testing components (NPC and monocular accommodation) and the performance of the total VVE compared with both our top factor model and the 5P prediction rule are important findings, demonstrating the potential of visio-vestibular deficits to augment the predictive power of symptom scores. Including VVE deficits makes the present study unique from the original 5P rule. Previous work has demonstrated the predictive power of visio-vestibular testing. Master et al,²⁷ in a group of 234 youth aged 5 to 18 years presenting to a concussion specialty clinic, found that accommodative amplitude strongly correlated with prolonged symptoms, with similar effect sizes found with VOR and saccadic eye movements. DuPrey et al¹¹ found a strong association between the initial NPC and prolonged recovery among 270 adolescents and young adults; however, in subsequent studies, the NPC has been found to be less strongly associated with prolonged symptoms compared with other elements of vestibular testing, possibly related to varying methodologies in measuring the NPC.^1,19 Evaluating the association of the vestibular/ocular motor screening (VOMS), a similar visio-vestibular assessment to the VVE, with PPCS has had mixed results. Among 79 college athletes, Whitney et al⁴⁵ found a strong association between the initial VOMS score and prolonged symptoms, whereas in younger adolescents, Knell et al¹⁹ found a more modest effect size, with an AUC of only 0.56 among women. One of the key differences between the VVE and the VOMS is the inclusion of monocular accommodation in the VVE.³ Even though the NPC 1 accommodation were included in 1 single factor in our study, it is certainly possible, based on the previous results from Master et al,²⁷ that accommodation is driving the predictive power of this factor, as accommodation contributes to the NPC. Although individually, in our study and in previous works, smooth pursuit, saccades, and VOR have shown predictive solid power, the fact that, as shown in Figure 3, they had redundant contributions suggests that only a subset of the VVE—specifically the NPC and monocular accommodation—might need to be added to an assessment of symptom burden to create the optimal predictive model.

Interestingly, we found that the point estimate of the AUC for the original 5P model in our data was lower than the unweighted model that included an adjustment for the varied time from injury in our population. There are several potential explanations for these differences. Our study enrolled exclusively from concussion specialty clinics; therefore, our participants might have had access to advanced treatments not necessarily available to the general population—such as targeted exercise or visio-vestibular therapy—enrolled in the 5P study. Also, as there are only a handful of concussion specialists practicing in the clinics, there likely is a smaller variety of interventions recommended compared with children enrolled from more diverse settings, such as with 5P. However, a study by Howell et al¹⁶ validating the 5P model in a concussion specialty clinic found similar diagnostic values between the original 5P study and specialty patients. Next, a difference between our study and the 5P study is the time from injury. On average, our participants were first evaluated 1 week from injury, compared with the 5P study, whereby all participants were enrolled within 48 hours of injury.⁴⁷ The improvement with adjustment of time from injury suggests that further out from injury, the 5P model may have less diagnostic value than in the acute time frame. Finally, the 5P model enrolled a much larger sample size, and our collection of data from clinical rather than scheduled research visits distinguishes the present study from the original 5P evaluation, as noted in the Limitations section.

Although age did not rise to the level of importance to warrant inclusion in our final model, it merits discussion as a predictor of PPCS. Among pediatric participants aged <18 years, older age has previously been found to be associated with prolonged concussion symptoms. In the original 5P study, one of the strongest effect sizes was found in adolescents (age, 13-18 years) compared with younger (age, 5-7 years) children.⁴⁶ We found our effect size to be in the opposite direction with age, with the younger children in our sample having the most prolonged recovery times. However, a closer look at our sample reveals that most of our participants were in the 13- to 18-year-old age group. Therefore, it is likely that younger adolescents (age, 13-16 years) are in the highest risk group for PPCS and that older adolescents (age, 17-18 years) begin to exhibit physiology that mimics their college-aged counterparts, who have been found to have shorter recovery periods compared with high school–aged youth.¹³ This hypothesis is supported by previous work from Corwin et al,⁸ who found that among 247 pediatric participants aged 5 to 18 years, the longest recovery times were in those aged 13 to 16 years, compared with those aged <13 years and those aged >16 years.

The critical need for risk stratification for PPCS in pediatric concussion is becoming more apparent as the evidence behind multimodal specialty interventions, and specifically, the success of early rehabilitation strategies, comes into focus. Multiple groups have shown that, among those seen in concussion specialty programs, those seen more promptly after injury have a lower incidence of PPCS.^21,34 The active management approach used in these specialty concussion programs, such as monitored aerobic exercise protocols and visio-vestibular rehabilitation, has independently been shown in randomized controlled trials to hasten recovery when administered early after injury.^20,22 Perhaps most excitingly, a recent clinical trial evaluating only youth at higher risk for PPCS (scoring moderate or high risk on the 5P scale) found a significant reduction in PPCS incidence for a group randomized to an aerobic exercise protocol versus usual care.¹⁶ Therefore, if we can successfully identify youth at the highest risk of PPCS acutely after the injury, we can provide targeted, early delivery of these critical interventions to the populations who will benefit the most from their initiation. A successful stratification approach is highly important given that these strategies can be time-intensive, may not be available to all youth, and may not be necessary for the 70% of youth who will not experience PPCS. Based on the present study, youth found to have high early symptom burden and abnormal NPC and accommodative amplitude would benefit most from early initiation of these therapeutics.

There are several limitations to our study. As our participants were primarily recruited from the concussion specialty program of our academic center, they may represent a cohort of children with a greater concussion burden—evidenced by the overall high rate of PPCS in the group when compared with previous studies.⁴⁶ Next, although we limited our sample to those seen within 14 days of injury, our mean time from injury to the first visit was approximately 7 days, and our data should not be extrapolated to the hyperacute time frame (eg, within 24 or 48 hours) after injury. A critical next step in this work is assessing the predictive power of symptoms combined with NPC 1 accommodation in a more acutely injured population (seen in more diverse clinical environments). As our study collected data from clinical visits, we were unable to determine the exact date of symptom return to baseline, which may have led to some misclassification of PPCS, with the exclusion of 46 participants for whom the PPCS status was determined to be unknown, as might have the utilization of obtaining preinjury symptom scales after injury due to recall bias.⁴² However, our sensitivity analysis in Appendix 2 (available online) showed that our modeling results are robust when considering the range of possible recovery status for these participants, and as most clinicians will not have a baseline symptom score available on the first evaluation, utilization of a preinjury symptom scale collected after injury improves the generalizability of our work. In addition, the clinical nature of our data led to a need for more standardization in the follow-up. We attempted to control for the varied follow-up times using kernel weighting of follow-up visits. Based on the results of the 5P study, we only included history elements of previous concussion history and migraine history⁴⁶; however, previous work has shown that other medical history elements, such as depression or anxiety, do have associations with PPCS.⁸ Finally, most of our participants were injured by a sport-related mechanism, while previous studies have found those injured by nonsport mechanisms. Specifically, motor vehicle collisions²⁴ and assault-related mechanisms²⁹ have protracted recovery trajectories. There may exist additional unique risk factors for prolonged recovery among these groups. Therefore, validation of our results among a larger sample, which includes both acutely injured children and those injured by diverse injury mechanisms, is warranted.

CONCLUSION

This study demonstrates that among commonly used clinical assessments for youth concussion, the vision testing component of the VVE can potentially augment symptom scores in predicting PPCS. Importantly, these findings can assist providers in determining which children and adolescents are at the highest risk for PPCS and might benefit most from early, active therapeutic interventions to improve recovery time and outcomes.