This study included patients within CHOP’s pediatric network, located in southeastern Pennsylvania and southern New Jersey, with more than a million annual visits. The network supports a socioeconomically, racially, and ethnically diverse population and accepts most insurance plans, including public insurance (eg, Medicaid). CHOP uses a single, unified EHR system for all aspects of care. We queried the CHOP EHR system to identify all individuals who had a visit to a network location for a concussion during the index period of July 1, 2012, through June 30, 2013 (n = 4977). Concussion visits were defined as those assigned an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis code of concussion; a full list of relevant ICD-9-CM codes has been published.²³ Patients were included regardless of previous concussion history. The population was then limited to patients who were 5–15 years old at the time of their first concussion-related visit in the index period (n = 3590; hereafter referred to as “index concussion”); the upper bound was selected to ensure that patients would still be seeking care at a pediatric practice for the entire 2-year follow-up period (with the latest possible date of repeat concussion of June 30, 2015). The population was further limited to active CHOP primary care patients (n = 2324), defined as those who had at least one primary care visit at one of CHOP’s 31 primary care offices within a 2-year period before their first index concussion visit; we expect that both previous and subsequent concussions for this group would be documented in the EHR regardless of where the concussion was diagnosed. Because the study necessitated in-depth EHR review, we generated a simple random sample of 25% (n = 577) for data abstraction. Distributions of race/ethnicity, sex, age, insurance payor, location of first visit (primary care, specialty care, emergency department/urgent care, hospital), and month of visit were similar for the selected sample and overall cohort (results not shown). A flowchart depicting selection of the study cohort is shown in the Figure.

A total of 8.4% (n = 45) of patients were diagnosed with a repeat concussion within a year of their index concussion; 16.2% (n = 87) had a repeat concussion within 2 years, including 3.4% (n = 18) who were diagnosed with 2 additional concussions. The median (IQR) time to diagnosis of a repeat concussion was 11.8 (5.8–17.8) months. The 2-year risk of a repeat concussion did not vary by sex or insurance payor (Table I). However, risk among 12-to 15-year-olds was 1.85 times that of 9-to 11-year-olds (19.8% vs 10.7%; 95% CI 1.09–3.13). Sixteen percent of all patients in the study had a history of concussion before the index concussion; 22.1% of these patients went on to experience a repeat concussion within 2 years of the index injury (compared with 15.1% without a history of concussion before the index concussion, P = .11). Overall, the 2-year repeat concussion risk was greater for patients who had ≥1 pre-existing co-occurring condition than those with no conditions (RR 1.49 [1.00−2.22]); more specifically, one-quarter or more of patients with a history of migraine/headache (28.6%) and anxiety (25.0%) had a repeat concussion.

The risk of repeat concussion by clinical characteristics of the index concussion is shown in Table II. Risk did not vary significantly either by the mechanism of injury of the index concussion or whether the injury was sports- or recreation-related. However, risk was particularly heightened among patients with a longer clinical course for the index concussion; the median clinical course was 17 days (IQR: 5–62) for patients with a repeat concussion compared with 11 days (IQR: 4–25) for those without a repeat concussion (P = .004). Furthermore, compared with patients whose clinical course was 0–7 days, those with a clinical course of ≥29 days were almost twice as likely to experience a repeat concussion (RR 1.92 [1.19–3.12]). Risk was also greater among patients with a greater number of reported PCSS symptoms during their index concussion: patients with ≥11 symptoms were more than 2 and a half times as likely to have a repeat concussion compared with patients with 0–2 symptoms (RR 2.66 [1.55–4.56]). Table III (available at www.jpeds.com ) shows the risks associated with specific symptoms. As there was strong correlation between the length of clinical course and number of PCSS symptoms for the index concussion (Spearman r = 0.59), likely because both are strong proxies for concussion severity, we constructed separate multivariable models for clinical course (Model 1) and symptoms (Model 2, Table IV). Both predicted increased risk. After we accounted for other factors, patients whose clinical course was ≥29 days had a 65% increased risk compared with patients whose course was 0–7 days (adjusted risk ratio [aRR] 1.65 [1.01–2.69]), and patients who experienced ≥11 symptoms during the course of care for their index concussion had over twice the risk of a repeat concussion compared with patients who had 0−2 symptoms (aRR 2.12 [1.21–3.72]). In both models, the presence of a co-occurring condition was not a significant predictor. In sensitivity analyses that excluded patients with documented concussions before the index concussion, aRRs for clinical course and symptoms were even stronger in magnitude (Table V; available at www.jpeds.com).

This study of patients at a large pediatric network quantified the risk of medically diagnosed repeat concussions among youth aged 5–15 and identify intrinsic and extrinsic characteristics that might influence that risk. Overall, we estimated that 1 in 6 youth, including 1 in 5 adolescents (ie, 12-to 15-year-olds), who were diagnosed with a concussion sustained at least 1 additional diagnosed concussion within 2 years. Combined with studies indicating that multiple concussions negatively affect both short- and long-term health and development, as well as the recovery process of subsequent concussions,^10–16 these findings suggest that a substantial proportion of youth who have concussions may be experiencing a high concussion burden, leaving them at particularly heightened risk for poorer outcomes.

We also identified several factors from the index concussion that predicted an increased repeat concussion risk, including symptom burden, length of clinical course of care, and patient age, although the latter may be related to increased sports- and recreation-related exposure. Although concussion grading is not currently recommended by any professional society, previous studies have shown that both symptom burden and length of clinical course are likely correlates of concussion severity.^24,29 Our data further extend these findings by suggesting that the level of concussion severity may be related to a greater risk of repeat concussion.²⁴ Further investigation is needed to define more precisely these categories so that providers can advise families regarding the “severity” of their index concussion and the likelihood that it leads to elevated risk for a future injury. Until then, healthcare providers may use the practical evidence regarding these factors as they counsel concussed patients and their families about their future risk of injury.

With respect to other factors, the presence of a pre-existing co-occurring condition, particularly migraine/headache, was found to be associated with an increased risk of repeat concussion in bivariate analyses. However, we did not find that co-occurring conditions predicted repeat concussion risk over and above symptom burden and clinical course in multivariable models, indicating that the concussion itself constitutes the main risk factor. Nevertheless, some studies of adult and adolescent patients with concussion have found that pre-existing ADHD, learning difficulties, mood disorders, and psychiatric conditions were associated with a life-time history of concussion and extended concussion recovery.^{11,12,19,30,31} Thus, it is important for providers to inquire about pre-existing co-occurring conditions, as this knowledge may influence the provider−family conversation about continued or future participation in sports with elevated risk for concussion. The interplay between these conditions and risk of future injury should be explored further in larger prospective studies.

Notably, risk did not vary by the mechanism of injury of the index concussion or whether that injury was sustained in sports and recreation. This reinforces the notion that all concussions, regardless of mechanism, contribute to an individual’s concussion burden and risk of subsequent injury; this finding also emphasizes the need for providers to discuss return to risk-bearing activity, such as contact sports, during follow-up concussion visits even when the index concussion was sustained outside of sports and recreation.

There are several potential limitations. First, collection of variables relied on patient data captured by existing EHR records. Thus, data were likely not compiled as systematically and consistently as they would have been if collected prospectively. However, CHOP introduced an EHR clinical decision tool in July 2012 that included a standardized template to systematically guide providers through concussion-specific assessments, diagnosis, and documentation of relevant information. Since introduction of the template, the vast majority of concussion visits have been documented using this format.³² In addition, our abstraction process involved both an electronic data collection tool and comprehensive review of each patient’s EHR to optimize accurate capture of relevant data. Regardless, some variables that may be important in predicting risk, such as whether the patient participates in contact sports, were not captured. Without these data, we could not examine whether the increased risk of repeat concussion observed for 12-to 15-year-olds was due to increased exposure to risk-bearing activities such as contact sports. In addition, there may also exist in our EHR-based study some ascertainment bias of those children who returned consistently to a CHOP network location as a patient. Second, although our previous study found that ADHD diagnosis is validly captured in CHOP’s EHR with a high sensitivity,³³ other co-occurring conditions, in particular those that are harder for primary care providers to diagnose, may not be as reliably captured. Third, the details in the EHR on repeat concussions managed solely by non-CHOP providers are likely limited, and we were not able to capture concussions where the patient did not seek medical care. We limited this study to CHOP primary care patients, as we expected these patients to be most likely to seek treatment at a CHOP location and that there would be more complete documentation in the EHR of all concussions sustained regardless of location of care for those patients. Nevertheless, this limitation likely resulted in an underestimated risk of repeat concussion. Fourth, the concept of “time to recovery” is difficult to capture, especially outside the sports setting in which return to play is commonly used as the outcome of interest. There is no clear consensus about how to define recovery. As a proxy, we used length of clinical course categorized into clinically meaningful categories in order to reduce potential misclassification bias.²⁴ However, some patients may experience lingering symptoms even after they discontinue clinical care. Forty-seven percent of study patients reported at least 1 PCSS symptom at their last visit; however, this did not differ for those with and without a repeat concussion. Finally, the population of CHOP primary care patients may not be entirely representative of the underlying population of the catchment area; the majority of patients were non-Hispanic white and had private insurance. Further examination of larger nationally representative samples would capture greater diversity and allow a more specific exploration of these demographic factors.

Given the potential effects of multiple concussions on health and development across the lifespan, including health consequences that appear later in life,³⁴ understanding concussion burden in childhood is critical. The current study demonstrated that one in six 5-to 15-year-olds who sustained an index concussion went on to experience a repeat concussion within 2 years. It also identified several factors, including age and markers of severity of the index concussion, which might influence that risk. This study provides clinicians with data to support recommendations to families regarding the need for effective prevention practices and risk management (eg, limiting contact sport participation) and informs public health conversations regarding efforts to prevent and mitigate the long-term effects of concussion. In coordination with recently released clinical guidelines for the diagnosis and management of pediatric concussions,^35,36 such discussions may lead to meaningful improvements in clinical care. Future longitudinal studies should be conducted to rigorously evaluate the short- and long-term consequences of these repeat injuries.