Study data and case definition
We retrospectively analyzed medical visits for pediatric concussions extracted from electronic health records (EHR) collected as part of routine clinical care for patients seen at Nationwide Children’s Hospital (NCH). NCH EHR uses a comprehensive and integrated set of clinical software systems to manage and record various patient care data domains such as demographics, medical visits, diagnoses, orders, and provider information that can be extracted to support data-driven research endeavors. For this study, we defined medical visits for concussions using the following International Classification of Diseases, Ninth and Tenth Revisions, Clinical Modification (ICD-9-CM and ICD-10-CM) codes: 850.0, 850.1, 850.11, 850.12, 850.2, 850.3, 850.4, 850.5, 850.9, and those beginning with S06.0. We extracted the following data elements from the EHR for each concussion: basic demographics (e.g., date of birth, sex, race) and medical visit information (e.g., diagnostic codes, date of injury, date and type of visit, date of first visit, date of symptom resolution, payor). Additional data regarding the total number of all-cause first medical visits during the study period were also extracted. This study was approved by the Institutional Review Board of the primary authors’ institution.
The population in this study was youth ages 10 to 17 years old who had a confirmed isolated concussion diagnosis, had medical visits for one or more concussions between July 1, 2012, and December 31, 2017 (not necessarily date of injury), and visited one of seven NCH concussion clinics at least once for their concussion treatment. Each medical visit for a unique concussion was determined by the date and time of the visit. Patients with repeated concussions were preceded by at least 90 days without an additional concussion diagnosis. We used the date of injury, date of the first visit, and date of symptom resolution to differentiate each unique concussion and included these in the analysis.
Patients were excluded if: 1) they were also diagnosed with a more severe TBI within 2 weeks of the initial concussion visit; or 2) they were receiving ongoing concussion treatment during the study period, but the patient’s first medical visit occurred before July 1, 2012.
A total of 5,211 initial concussion visits were identified, with 4,955 meeting criteria for analysis. Figure I presents the exclusion process for initial visits to arrive at the final sample used in the analysis.
Study variables and measures
Rates of initial concussion visits were calculated as the number of first medical visits among youth with concussions in a year divided by the total number of first medical visits of NCH patients in the same year, then multiplied by 10,000.
Type of initial concussion visit was classified as 1=Sports Medicine; 2=ED, including ED or urgent care centers; and 3=Other, for all other specialties.
Days from injury to initial concussion visit were measured as the number of days from date of injury to date of first medical visit.
Insurance type was measured based on insurance plans across all medical visits associated with each injury (1=Public (e.g., Medicaid), 2=Private, 3=Self-pay, and 4=Other). Self-pay and other were later collapsed together. Injuries with differing insurance plans across medical visits were labeled as other.
Other variables included patient demographics (e.g., age, sex, race), season (quarter 1 to 4) and calendar year of medical visit (calendar year from 2012 to 2017).
We described and compared demographic and injury characteristics of the study participants across different insurance types using chi-square tests. We examined trends across years and their potential interaction with insurance type via statistical models. Specifically, we employed Poisson regression to model the rate of initial healthcare visits for concussions versus initial visits for all causes, logistic regression to assess the proportion of concussion-related initial visits to sports medicine at NCH (or to ED) vs. otherwise, and linear regression (with log-transformed outcome) to determine the number of days from injury to initial visit to NCH. In all regression analyses listed above, we used hierarchical modeling to account for a patient with multiple concussions by nesting these injuries within a patient and assessed the interactions between year and insurance type. Further, we adjusted for study year and season, patient age, sex, and race (patient race was not included in the Poisson regression analysis since denominator information was not available at that level). Data were analyzed using SAS (version 9.4, SAS Institute Inc, Cary, NC), and a statistical significance level was set a priori for each test at α=0.05.