Methods
The Florida Agency for Health Care Administration (AHCA) ED and inpatient data sets from 2010 to 2014 were used in this analysis. The data sets are mutually exclusive, so ED patients discharged into the inpatient unit of the same hospital are not included in the ED data. The data include demographic variables, up to 30 diagnoses, and external cause of injury codes (E-code) for patients who had an ED visit or admission to an acute care hospital. The AHCA also releases annual hospital financial data, which include ownership status, location and financial information. The hospital factors were merged with the patient data for each year so the model could control for differences in the 123 Florida hospitals.
Inpatient and ED patients between the ages of 5 and 18 who had a sports-related E-code were included in the analysis. Patients were categorised into age groups approximating various school divisions: elementary school included ages 5–10, middle school included ages 11–13 and high school included ages 14–18. Other patient demographics such as gender, race, ethnicity and payer type were used in this analysis. Payers are the insurance companies who pay for patients’ healthcare. This study included the three main types of insurance for youth in the USA, which are commercial (private insurance companies), Medicaid (state-run insurance for youth in low-income families) and uninsured (the patient is responsible to pay their healthcare costs). Payer types and status are included in the model to capture potential differences in utilisation related to comprehensiveness of coverage and, by extension, the out-of-pocket price to the patient. For example, the uninsured may seek less care as a way to avoid paying full price at the time of delivery. Conversely, commercially insured patients may request more services as their out-of-pocket price at the time of delivery is reduced to copayment arrangements. The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) Injury Severity Score (ICISS) method was used to measure injury severity. ICISS ranges from 0 to 1, with unity indicating 100% survival and 0 implying 100% mortality. The lower the ICISS, the more severe the injury or combination of injuries. The severity variable used was ICISS multiplied by 100 in order for the model estimates to be more easily interpreted.
Patients who had an injury from a sport were identified using the following E-code fields: E006.x (individual sports), E007.x (team sports), E008.x (other sports), E886.0 (fall from sports), E917.0 (struck in sports) and E917.5 (struck and fall in sports). These include all ICD-9-CM codes that had ‘sports’ in the description. The inpatient data included 4658 observations and the ED data included 234 754 observations. Observations were omitted from the model analysis if they did not include an E-code for a specific, named sport, for example patients who were injured with an E-code of a general nature such as ‘struck in sports’ or ‘other activity involving other sports’. Observations were also omitted from the model analysis if the patient did not seek treatment for one of the injuries defined in the Barell Injury Diagnosis Matrix. The Barell Injury Diagnosis Matrix is a commonly used tool in injury epidemiology that uses ICD-9-CM codes to classify injury by body region and nature of injury. Examples from those omitted observations included youth patients who were principally diagnosed with an unspecified episodic mood disorder or other cellulitis or abscess. One patient whose costs were 12 times higher than the average was omitted as a cost outlier. For the final analysis, the model included 2303 inpatient observations.
The sports E-codes were categorised according to the American Academy of Paediatrics’ Committee on Sports Medicine and Fitness (2001).21 The categories were full contact or collision sports, limited contact sports, and non-contact sports. The full contact sports group included observations with E-codes of E007.0 (American football), E007.2 (rugby), E007.4 (lacrosse/field hockey), E007.5 (soccer), E007.6 (basketball), E008.0 (boxing), E008.1 (wrestling) and E008.4 (martial arts). The limited contact group included E006.0 (roller-skating/skateboarding), E006.1 (horseback riding), E006.4 (bike riding), E007.1 (flag football), E007.3 (baseball), E007.7 (volleyball), E008.2 (racquet/hand sports) and E008.3 (Frisbee). The non-contact sports group included observations with E-codes of E006.2 (golf), E006.3 (bowling), E006.5 (jump roping) and E006.6 (non-running track and field).
The principal diagnosis code of the patients was used to create the nature of injury categories according to the Barell Injury Diagnosis Matrix. Injuries were categorised using the matrix into fractures of the skull, neck and trunk; other fractures; sprains and strains; internal; open wound; amputations; blood vessels; contusion/superficial; crush; burns; nerves; and unspecified according to the principal diagnosis code of the patient.22 The reference group for the analysis included sprains and strains and contusion/superficial injuries. Burns, blood vessels, nerves, amputation and crush each accounted for well under 1% of the total observations. Therefore, these were added to the unspecified injury observations and this variable was called ‘other injuries’.
The inpatient cost model was analysed using a linear regression with residence county fixed effects, meaning the variables were analysed within each county to control for differences between counties. The dependent variable was cost of the hospital visit. This was calculated from the total charges of the visit as reported in the AHCA. The total charges were multiplied by each hospital’s annual weighted cost-to-charge ratio to estimate the actual patient care cost. Cost-to-charge ratios are the reported total costs divided by the total revenue of each cost centre. Cost-to-charge ratios were calculated for each hospital for each year. The cost centre ratios were then combined for an annual weighted overall hospital cost-to-charge ratio. The costs found were then adjusted for inflation to 2014 dollars using the producer price indexes for hospital inpatient care and hospital outpatient care accordingly. The distribution of costs was highly skewed; therefore, the cost variable was transformed using the natural logarithm. Microsoft Excel 2016, Microsoft Access 2016 and SAS V.9.4 software were used in this analysis.