|
Abstract
|
Published research1,2 suggests the COVID-19 pandemic exacerbated already increasing rates of overall US firearm mortality. Descriptive comparisons of 2019 to 2020 mortality data suggest this early trend was driven by homicides and urban firearm deaths.2 However, to our knowledge, no study has quantified excess intent- and urbanization-stratified US firearm deaths throughout the pandemic using inferential statistical methods. We estimate overall and intent- and urbanization-stratified excess US firearm mortality throughout the COVID-19 pandemic.
In this cross-sectional study, monthly US firearm mortality data for 1999 to 2021 (most recent) were extracted from CDC WONDER.3 Data on race and ethnicity were not collected because they were not used in the hierarchical time series modeling or forecasting approach. CDC WONDER also does not report mean age. Urbanization- and intent-stratified mortality were aggregated using a hierarchical time series structure, with urbanization- and intent-stratified deaths aggregated to intent level and intent aggregates combined into a final aggregate time series. Urbanization-level forecasts were produced by aggregating urbanization- and intent-stratified deaths to the urbanization level. The West Virginia University Institutional Review Board deemed this study exempt because it did not meet the definition of human subjects research. This study followed the STROBE reporting guideline.
A counterfactual scenario was created in which US firearm mortality continued prepandemic trends. Counterfactual mortality estimates were created by forecasting 21 months of pandemic-era (April 2020 to December 2021) data using an ensemble time series model of prepandemic (January 1999 to March 2020) data. This approach combines forecasts of multiple time series models and has improved accuracy over single-model approaches.4 Our ensemble model pooled autoregressive integrated moving average and exponential smoothing forecasts using inverse variance weighting. Minimum trace forecast reconciliation was used to integrate the unique influence of each hierarchical time series level into final aggregate forecasts.
During the pandemic, 83 966 firearm deaths occurred vs 73 194 predicted deaths, representing a 14.7% increase above expected (99% prediction interval [PI], 13.6%-15.9%) (Table). Stratified by intent, firearm homicides showed the greatest increase above expected values (9329 additional deaths [34.1%; 99% PI, 31.4%-36.8%]), followed by other (327 deaths [13.2%; 99% PI, 9.0%-17.7%]) and suicide (1116 deaths [2.6%; 99% PI, 1.2%-4.0%]). Larger increases were observed in urban areas, and homicides showed the largest percentage increase for each urbanization category. Monthly homicide firearm deaths were higher than forecasted estimates throughout the pandemic; firearm suicides were not higher than forecasted until later in the pandemic...
Language: en |