Study Design

Institutional review board approval was obtained, and all participants (or their guardians) provided written, informed consent for research participation. Our study was a retrospective, case-control design. Participants were recruited into the Vanderbilt Sports Concussion Center (VSCC) database from high schools in the middle Tennessee area that had participated in regional neurocognitive testing programs during 2007–2013. Most high schools in this region, which represents a diversity of socioeconomic demographics, school size, and geography, participate in this program.

Patient Data Identification

Patients with PCS after SRC were identified from the VSCC database. After a child experienced head trauma, a certified athletic trainer or team physician diagnosed concussion if the examiner noted the following on-field or sideline signs or symptoms: 1) lethargy, fogginess, headache, and so on; 2) altered mental status; 3) loss of consciousness; and/or 4) amnesia. Following the recommendations of the Concussion in Sports Group consensus guidelines, we used no grading system for concussion severity. All peri-injury consultation notes, vital signs, hospitalizations, radiological images, outside medical records, and clinical communications were reviewed in our electronic medical record.^45,46

Patients with PCS were defined as those experiencing postconcussion symptoms for greater than 3 months. Control patients were defined as those with documented symptom resolution by 3 weeks. The inclusion criteria were as follows: 1) the patient sustained the index concussion while playing a sport and 2) the patient was 9–18 years of age at time of injury. Exclusion criteria were as follows: 1) symptoms persisted from 3 weeks to 3 months and 2) no verifiable documentation of symptom resolution was available. Various studies diagnose PCS in patients with symptoms lasting 1 month; however, we chose to adhere to the strict temporal definition in the Diagnostic and Statistical Manual of Mental Disorders, Edition 4, which states that PCS patients are symptomatic beyond 3 months.^1,2,27,39,59

During 2011–2013, a total of 1116 patients were seen at VSCC. From this group, 40 patients identified as having PCS resulting from a sports-related mechanism were identified from a cohort referred to the Pediatric Neurology Clinic at Monroe Carell Jr. Children's Hospital at Vanderbilt. The first 40 patients with PCS were identified in alphabetical order and then matched by both age and sex to 2 control patients (1:2 matching) who had sustained an SRC but who had clear documentation by a trained health care provider (those with an MD, DO, PhD, PA, or NP degree) of symptom resolution, at rest and with exertion, within 3 weeks of injury.

Data Collection

From each patient's electronic medical record, we compiled provider consultation notes, radiological findings, medications prescribed, clinical communications, and scanned documentation. We collected data in 4 categories: 1) demographic variables; 2) key past medical, psychiatric, and family history; 3) acute-phase postinjury symptoms (at 0–24 hours); and 4) subacute-phase postinjury features (at 0–3 weeks). Significant stressors were noted if they represented major life events that were acknowledged by the patient or provider as possibly interfering with concussion recovery and symptomatology.

Although in a retrospective study it is difficult to independently corroborate medical, psychiatric, and family history, we used all peri-injury consultation notes, hospitalizations, outside medical records, and clinical communications available in our robust electronic medical record to confirm these variables. Most patients were queried by use of a standard health history form, affording each the opportunity to self-report medical, psychiatric, and family history. For symptoms in both phases, we looked for categorical endorsement of symptoms on the widely used Post-Concussion Symptom Scale (PCSS).^34,36 However, because of variability in the format of provider documentation of symptom endorsement, we were not able to use the 0–6 scale of the PCSS; instead we assessed the categorical endorsement of a particular symptom (endorsed or not endorsed). Symptom clusters are detailed in Table 1. Total scores for endorsement of representative symptoms in each cluster were calculated. These data were compiled into a single database and analyzed. Body mass index (BMI) data were collected if they represented measurements taken within 6 months of the SRC.

Statistical Analyses

Descriptive statistics are reported as mean ± SD for continuous variables and as frequency and proportion for categorical variables. Before performing analyses, we assessed variable distributions for normality by using histogram and Kolmogorov-Smirnov statistics. Univariate and bivariate association analyses were performed for demographic variables, presenting characteristics, and health care utilization variables to describe their distributions and assess their association with PCS, respectively. The chi-square Fisher exact test was used for categorical variables, ANOVA was used for normal continuous variables, and the Mann-Whitney U-test was used for nonparametric continuous variables. For bivariate analyses, significance was determined at a level of α = 0.05.

To identify the predictors most strongly associated with development of PCS, we used forward stepwise binary logistic regression models (P_in = 0.05, P_out = 0.10). Each model was controlled, a priori, for patient age, sex, number of previous concussions, and race. Acute and subacute symptom cluster scores were entered into the model a priori according to our hypotheses, namely that the number of symptoms would be higher among PCS patients than among control patients for each of the 4 symptom clusters (somatic, cognitive, sleep, and emotional). Bivariate associations between PCS and independent variables were analyzed. Variables found to have a trend-level association (p < 0.100) with PCS were assessed for collinearity by using the Spearman rank correlation coefficient. When collinearity was found, the variable with the weaker association with PCS, as defined as a smaller absolute correlation coefficient, was disqualified from the list of candidate predictor variables entered into the regression model. This regression model was evaluated for assumptions and aptness. Significance for the logistic regression was set at 0.01 (calculated as α = 0.05/5) via the Bonferroni continuity correction; 5 final variables were entered into the stepwise model. To generate the specificity and sensitivity data, we developed this model and then applied it retrospectively to the same data set. Statistical analyses were performed by using SPSS Statistics, version 20.0.0 (IBM Corp.).

Demographics

Characteristics of the 2 groups are detailed in Table 2. Several potentially confounding factors such as age (p = 0.722) and sex (p = 0.848) did not differ significantly between groups. Regarding the sport associated with the concussion, we found no significant difference (p = 0.197) between the 2 groups in the 5 major involved sports (football, basketball, baseball/softball, soccer, and other). The 3 most common sports categorized as “other” were equestrian sports, water sports, and lacrosse.

Neither race (p = 0.134) nor type of insurance (p > 0.999) was associated with PCS. Female sex did not predict any symptom cluster (somatic, cognitive, sleep, or emotional) in either group. When analyzing BMI data measured within 6 months of injury, we found no significant difference in the percentage of children in whom PCS did (69.3%) and did not (72.6%) develop (p = 0.480).

Medical, Psychiatric, and Family History

One possible confounding factor that was purposely not kept constant was the number of previous concussions, because the literature indicates that previous concussions represent a major potential risk factor for PCS.⁷⁰ In this cohort, athletes with PCS reported having previously sustained more concussions than did controls (p = 0.010). At least 2 previous concussions had been sustained by 27.5% of PCS patients and only 7.5% of control patients. PCS was more likely to develop in athletes with history of premorbid mood disorders (p = 0.002) and psychiatric illness (p = 0.039) but not migraine (p = 0.088). Significant stressors (close family member deaths and bullying) during SRC recovery was reported by 3 (7.5%) PCS patients. Other PCS predictors were a family history of mood disorders (p = 0.006) and migraine (p = 0.003). No significant association was found between development of PCS and presence of attention deficit hyperactivity disorder or a learning disability (p = 0.750).

Acute and Subacute Symptoms

No bivariate associations were found between acutephase symptom scores or subacute-phase symptom scores and development of PCS. Delayed symptom onset, defined as the report of being asymptomatic for at least 3 hours postinjury, was 10 times more prevalent among PCS than control patients (25.0% vs 2.5%, p = 0.001). PCS was not predicted by loss of consciousness, amnesia, use of an over-the-counter pain reliever, use of a narcotic pain reliever, initial consultation with a healthcare provider, or wearing a helmet.

Overall PCS Prediction Model

Bivariate analysis identified a personal or family history of mood disorders and the presence of delayed symptoms as candidate predictor variables to be entered into the forward stepwise binary logistic regression model. For model entry, the forward stepwise logistic regression model used these candidate variables along with acute and subacute symptom cluster scores, history of mood disorders, family history of mood disorders, presence of delayed symptoms, and the acute emotional cluster. The overall model explained a significant proportion of variance in PCS development with the Nagelkerke R² = 0.450 and was significant at p < 0.001 according to the model chi-square statistic. The model predicted 80.2% of correct classifications overall with a sensitivity of 55.0% and a specificity of 92.6%. Independently, a history of mood disorders was associated with increased risk for PCS development (relative risk 17.9, 95% CI 2.9–113.0, p = 0.002), along with presence of delayed symptoms (relative risk 20.7, 95% CI 3.2–132.0, p = 0.001). The acute emotional cluster scores were not associated with a decreased likelihood of PCS when the Bonferroni-corrected significance level of α = 0.01 was used. Results of the final logistic regression model are detailed in Table 3.

Demographics

We found that for this cohort, insurance status did not predict PCS. Although evidence suggests that high family stress in higher-functioning families with greater environmental resources may predict PCS in children with mTBI, our data do not indicate an effect of having private insurance over state or federally funded programs.^56,69

We also did not find an association between race and development of PCS. This finding was notable because children of minority race are significantly more likely to experience underdiagnosis; undertreatment; and conditions like asthma, attention deficit hyperactivity disorder, and learning difficulties.^{14,20,21,54,67} In addition, a prospective, observational study of 71 adults with mTBI who sought care at an emergency department found that although patients were discharged with instructions to follow up with a primary care provider within 1–2 weeks, African Americans were less likely to do so (OR = 0.36, 95% CI 0.13–0.99).⁴ In that study, patients without a primary care provider were assigned one, which introduced the patient to an unknown physician. Also in that study, confidence intervals were remarkably wide. Because that study examined adults with all forms of mTBI, it is not clear if this trend would be expected for children with SRC.

Medical, Psychiatric, and Family History

We found that PCS was more likely to develop after concussion in young athletes with preexisting mood disorders or psychiatric illness. These findings corroborate those of studies of general mTBI in older patients, that PCS is more likely to develop after mTBI in adults experiencing depression, anxiety, and/or life stressors.^{16,29,41,57,68} However, whether this finding applies to young athletes with SRC remains unclear. One prospective study of 130 children 6–15 years of age who received care for mTBI at 2 emergency departments were followed up at 1 week and 3 months after injury.⁵⁹ Concussions were sustained mostly from falls (34%), cycling accidents (21%), and sports (24%). Of the 85% examined during a 3-month follow-up visit, PCS had developed in 17%. Those with PCS were more likely to have had a previous head injury (p < 0.001), learning difficulties (p = 0.02), psychiatric illness (p = 0.02), or premorbid family stressors (p < 0.001).⁵⁸ One major limitation of the aforementioned study was that postinjury behavior and symptoms were obtained from the parent or guardian, not the child. If parental stress modifies endorsement of PCS symptoms, it is unclear if these data truly reflect the child's symptoms.⁵⁶

Prevalent among our cohort of young athletes with PCS after concussion were family histories of mood disorders (p = 0.006), other psychiatric illness (p = 0.031), and migraine (p = 0.003). It is worth mentioning that migraine histories were determined on the basis of self-report of a migraine disorder or chronic headache syndrome. For example, many patients reported that siblings at times experienced “headaches.” However, we only included patients who self-reported either a significant chronic headache syndrome or migraine disorder.

This higher prevalence of mood disorders and psychiatric illness is consistent with previous findings that preinjury parental anxiety and family and life stressors predict protracted mTBI recovery in children.^56,58 These findings emphasize the value of addressing concussion recovery specifically in children and adolescents, who must recover within the context of complex modifiers like parental stressors. It is difficult to say whether these proposed modifiers are genetic, environmental, or both. True manifestations of PCS can be difficult to distinguish from symptoms of primary disorders of depression, anxiety, and migraine.^10,48 Baseline health and psychiatric status of both the child and the primary caregivers should be considered with regard to symptom reporting and expectations regarding recovery.

Acute and Subacute Symptoms

We did not find that loss of consciousness, amnesia, or any symptom cluster predicted PCS. This finding probably results from the fact that these risk factors have been identified in youth cohorts that included all forms of mTBI, not just those resulting from sporting activities. These more severe forms of impact include motor vehicle collisions, falls, and assaults. However, one unexpected finding was that in our cohort, PCS was significantly more likely to develop in athletes who endorsed delayed symptoms (onset > 3 hours postinjury) (p < 0.001). A common scenario was the athlete continuing to play after the significant hit and only noticing symptoms well after the game. The implication of this finding is that athletes who do not experience symptoms immediately might not be removed from play at the actual time of injury. This implication is especially true given that loss of consciousness and amnesia, obvious and severe manifestations of SRC, were rare in our cohort and in the literature for this population.⁴⁹ By remaining in play, these young athletes may then experience second hits, exposing their already injured brain to additional insults. The interaction between delayed symptoms and development of PCS is complex and warrants further study.

Second hits in a short time frame are significant for 2 reasons. First, the American Academy of Pediatrics, the American Medical Society for Sports Medicine, the American College of Sports Medicine, the Concussion in Sports Group, and the American Academy of Neurology have all issued statements instructing both physical rest and cognitive rest immediately after SRC in a young person, based on evidence that rest expedites symptom-free recovery.^{22,25,26,28,45} However, although recommended by these groups, the exact nature and quantifiable benefit of quality physical and cognitive rest remains controversial.^12,54,55,67 Second, decreasing the time between repeat concussions is an independent risk factor for protracted recovery in mice and children.⁵² Beckwith et al. and Duhaime et al. highlighted the difficulty of SRC diagnosis because of variability in on-field symptoms, an athlete's willingness to report, and the potential for delayed symptom onset, seen in up to 50% of a sample of college football players with SRC.^8,17 Furthermore, subconcussive impacts are increasingly being recognized as having potentially negative long-term cognitive effects for those who play contact sports. Talavage et al. reported that for 11 male football players who had no clinical concussion symptoms over the course of 1 season, neurocognitive testing indicated development of significant deficits, highlighting the risks and ramifications of repetitive, subconcussive impacts.⁶⁵ Another study by Duhaime et al. used instrumented helmets to follow 450 college football and ice hockey teams. Although 486,594 head impacts were recorded during the study period, for one-third of all diagnosed concussions, the contact event was not clinically apparent to officials or by report from the athlete.³²

In another prospective study of 1208 college football players wearing instrumented helmets, concussions were more likely to be diagnosed immediately after impacts with the highest kinematic measures. These more forceful hits resulted in these same players being removed from play immediately. However, those players for whom diagnosis was delayed were more likely to continue to play and have a higher number of recorded, subconcussive, repeated head impacts.⁸ These findings could potentially explain why young athletes with SRC in this study in whom PCS later developed were more likely to report delayed symptom onset; if symptoms were delayed, diagnosis would be delayed, play would continue with risk for repeated impacts, and beneficial physical rest and cognitive rest could not begin. Although each of these studies focuses on older, collegiate athletes, there is no plausible reason to think that the younger athletes in our study are less prone to these same repetitive, subconcussive impacts.

All patients included in this study experienced a “big impact” during play, based on the medical record history provided by the patient or a family member. We offer only serial hits or lack of rest as possible explanations. However, other possibilities could explain the connection between delayed symptoms and PCS. The high prevalence of behavioral health problems in the PCS cohort may also influence the onset of symptoms. Regardless, these preliminary data argue for prospective evaluation of serial hits and delayed symptom onset in larger samples.

Limitations

The lack of predictive symptom clusters could be explained by a limitation in our study, which precluded the use of anything beyond categorical variables for endorsement of a given symptom. Although we used symptoms present on the PCSS, because of variability in formats of symptom documentation by providers, a 0–6 scale was impossible to use consistently for patients in this retrospective case-control study. Widely used standard forms like the Rivermead Post-Concussion Symptoms Questionnaire and the PCSS have severity of symptom scales, which allow more nuance for the endorsement of a given symptom.^13,32,34 However, these forms were not consistently used by clinicians involved with patients in our study. Our observed variation in symptom documentation used by providers is troubling and is probably a source of significant variability in the literature.

We also do not know what the scores on such a scale would have been for patients in this cohort before an SRC. This baseline information would be useful, given the relative nonspecificity of postconcussion symptoms. This finding speaks to the urgent need for consistent provider use of a symptom scale to allow quality continuity of care between emergency departments, primary care physicians, sports medicine providers, and other specialists. Additionally, it is unclear if symptom scales currently in use are generalizable to populations at increased risk for PCS, like those with mood disorders, migraine histories, or numerous previous concussions. Both the retrospective nature of our study and its location in only 1 region of the country limit the generalizability of our results.

Future Directions

The lack of verified biological explanations for the nonspecific symptoms seen with PCS has led many to question its legitimacy as a disease entity.^18,48 As a result, many have recently focused efforts on identifying biomarkers that might predict or be hallmarks of PCS development. Noting that the G(-1019) allele of HTR1A is associated with major depression and suicide, a team led by Smyth et al.⁶⁴ examined the prevalence of the G(-1019) allele in children with mTBI. In their cross-sectional study of 47 symptomatic children who experienced postconcussive symptoms for 7 or more days, the G allelic frequency and genotypic frequency for HTR1A polymorphisms was similar to that among controls.⁶⁴ In another study, S100B, an astroglial calcium-channel binding protein was found to be elevated after mild to severe TBI.¹ S100B is also highly correlated with abnormal cranial CT scans of mTBI patients; sensitivity is remarkable (90%–100%).^1,2 A retrospective analysis of 76 children with mTBI measured S100B levels immediately after mTBI; however, no association was found between serum levels of S100B at this time and later development of PCS.¹ Cleaved tau protein, although elevated in patients after intracranial injury and correlated with functional outcomes after moderate to severe TBI, is also not detected more often in adult patients with PCS than in those without.^7,9,31,39 However, because each of these studies in this developing field focused on head injury of adults, the current generalizability to SRC in youth remains unclear.