Accurate data on firearm injuries in the United States are critical to better understand the total cost or societal burden of gun violence. However, there is no single resource that provides reliable estimates of this type of injury in that country. Data are scarce and, in some cases, too expensive for most researchers.

This lack of information limits the ability of researchers to answer basic questions about gun violence, such as whether injury trends are changing over time or whether harm reduction strategies are effective.

As part of the Gun Policy in America initiative, RAND researchers developed a longitudinal database for the of U.S. national estimates of hospitalisations for non-fatal firearm injuries between 2000 and 2022.

These estimates are based on a variety of data sources, the largest of which are summaries from state inpatient databases (SIDs), which are supplemented by other data from the country’s health department. The estimates do not include emergency department visits that do not result in subsequent hospitalisation or firearm injuries for which inpatient medical care is not sought.

Between 2000 and 2022, there were an estimated 728,617 hospitalisations for non-fatal firearm injuries nationwide: that is an annual average of about 31,600 non-fatal injuries in the United States.

The data also reveal substantial differences in the rate of non-fatal firearm injury hospitalisations among states and relatively stable national trends. These data could help answer questions regarding the effects of state policies and the social cost of firearm injuries to states and the country.

Data and methodology

The model used for non-fatal firearm injury hospitalisations combined SID data that were available through the Healthcare Cost and Utilization Project (HCUP), comprehensive state SID data, data available through U.S. health department web portals on hospitalisation, and data provided by state health departments in response to direct requests from the study team.

Years with data lacking the number of firearm injury hospitalisations were imputed using hypothesised covariates associated with firearm hospitalisations. These covariates included the non-suicide firearm death rate, the rate of firearm deaths in medical facilities, the rate of reported violent crime, and annual demographic, economic, and social characteristics at the national level.

Both the imputation of missing hospitalisation data and the correction for incomplete injury mechanism data were estimated simultaneously in a Bayesian regression model. This model uses a complex error structure that was designed to capture key features of the hospitalisation data, including the greater variance in firearm hospitalisation rates for less populated years and the correlation of values within each state over time. Multiple imputation datasets were created by sampling from the posterior predictive distribution of the model.

_____

Aquest apunt en català / Esta entrada en español / Post en français