Our study used administrative discharge data from the 1993–2008 Nationwide Inpatient Sample (NIS) of the Healthcare Cost and Utilization Project (HCUP), sponsored by the Agency for Healthcare Research and Quality (AHRQ), Rockville, Maryland. The NIS is created by AHRQ from the State Inpatient Databases provided by public and private statewide data organizations from participating states as part of a federal-state-private collaboration. The NIS is the largest, longitudinal, all-payer inpatient care database in the United States, with an average of 8 million hospitalizations from ~1000 hospitals each year. The NIS approximates a 20% stratified random sample of all short-term US community hospitals. The sampling frame for the 2008 NIS includes State Inpatient Databases from ~95% of all hospital discharges in the United States. Data were obtained from the HCUP Central Distributor, with approval from the institutional review board at the Johns Hopkins University. The NIS includes both patient level data such as demographics (eg, age, gender, race), admission type and source, up to 15 ICD-9-CM diagnostic and procedure codes, insurance status, total hospital charges, length of stay, discharge disposition, and hospital level information (eg, hospital ownership, number of beds, urban/rural, geographic region, and teaching status).
For this study, eligibility was limited to children who were ages 0 to 19 years at admission and who were hospitalized with a primary or secondary International Classification of Diseases, Ninth Revision, Clinical Modification
(ICD-9-CM) diagnosis code for drowning injury (994.1). Patients who died while hospitalized were included. The NIS contains no unique identifiers, so to reduce the effects of double counting multiple hospitalizations for the same drowning-related injury, the hospitalizations of patients who were discharged to another short-term care hospital were not included in the analysis (ie, only the terminal hospital admission was included).12
This assumes that these cases would likely be captured at the point of the definitive drowning care in the receiving hospital records.
Circumstances of drowning were determined based on the external cause of injury code (E-Code). We categorized circumstances of drowning injury into 5 groups: recreational swimming and diving (E910.2–910.3), in bathtubs (E910.4), other drowning activities (E910.0, E910.1, E910.8, E910.9), all other E-codes, and missing (E-code not available). Similarly, we determined intent of injury (ie, unintentional, intentional, undetermined) by using the available E-codes.
To generate national estimates of hospitalizations from the NIS, we used the appropriately scaled discharge weights provided by HCUP.13
With these weights, national estimates of hospitalization rates are rendered comparable across years, despite the varying number of states participating in each year of the HCUP project. All analyses were performed with the appropriate weighted stratified sample design, using the survey command (SVY) options within Stata 10.0/MP (College Station, TX). For the incidence rate calculations, we used US Census estimates for the national civilian population at midyears during this time interval.14
We present estimated crude hospitalization incidence rates by age, gender, and outcome. Because E-codes were missing for up to 55% of hospitalizations before 1997, we compared 2-year aggregate data for years 1998–1999 and 2007–2008 (where codes were missing for only ~10% of hospitalizations) to evaluate changes in drowning mechanism and intent over time.