Identifying and Characterizing Firearm Storage Practices; Causal Links to Veteran Suicide

About This Grant

Significance to VA: Firearm suicide prevention has been prioritized by VA clinical, operational, and research leadership. To date, VA firearm suicide preventions have almost entirely focused on promoting in-home secure firearm storage as a means to increasing the time it would take for an individual with suicidal intent to access a loaded firearm, but evidence to guide delivery of these interventions to the Veteran population is lacking. Innovation and Impact: The proposed effort is innovative in that investigators will use a natural language processing (NLP) approach (e.g., large language models) to identify Veterans with firearm access and characterize their specific firearm storage practices, and link to data from the National Death Index to accurately identify mortality events and provide rich information about patient cohorts. This work will provide critical insights into the potential benefits of different secure firearm storage behaviors and identify patient populations who might benefit most from them. Specific Aims: Aim 1: Develop an NLP approach to identifying and characterizing specific firearm storage practices among VHA-using Veterans with firearm access, and compare model performance among key subgroups. Aim 2a: After identifying a cohort of VHA-using Veterans with firearm access and characterizable firearm storage practices, we will describe and compare differences in storage practices among demographic and clinical subgroups, overall and stratified by whether Veterans have clinically underappreciated suicide risk. Aim 2b: Identify the patient, clinical, and setting factors associated with Veterans having discrepancies in described firearm access and storage practices between their structured data (e.g., CSRE, SSP) and unstructured data (regular note text). Aim 3: Perform causal analysis of the relationship between specific storage practices and firearm suicide mortality among Veterans with firearm access, including among subgroups (e.g., underappreciated risk). Methodology: In Aim 1, the research team will engage subject matter experts to develop a firearm storage term list and test several NLP approaches, including use of large language models, to identify and characterize firearm storage practices among those with firearm access. Model performance will be tested among key demographic (e.g., sex, race) and clinical (e.g., clinically underappreciated suicide risk) subgroups. In Aim 2a, the research team will develop a cohort of VHA-using Veterans with characterizable firearm storage practices among the national population of Veterans, and describe differences between those using secure and less- secure types of firearm storage. In Aim 2b, the cohort will be merged with existing firearm-related structured data elements gathered during completion of Suicide Safety Plans or Comprehensive Suicide Risk Evaluations to identify and describe circumstances in which Veterans have discrepant firearm data. In Aim 3, the cohort created in Aim 2a will be linked to mortality data to model the association between specific firearm storage practices and risk of firearm suicide mortality. Negative control outcomes will be used to assess for residual confounding in results. Age-adjusted firearm-specific homicide and unintentional injury death rates will also be described among the cohort. Path to Translation/Implementation: Findings from this study are expected to provide evidence that guides a wide variety of VA suicide prevention efforts, as well as ongoing efforts in the Department of Defense, non-VA healthcare systems, and community settings.