We used the most recent available data from the National Violent Death Reporting System (NVDRS) Restricted Access Database to compare victim, perpetrator, and incident characteristics by homicide mechanism (firearm vs. non-firearm) among older adults, and Fatal Injury Data from Web-based Injury Statistics Query and Reporting System (WISQARS) of the Centers for Disease Control and Prevention (CDC) to examine trends of these homicides over time. We defined ages 60 years or older as “older adults” as per the Uniform Definitions and Recommended Core Data Elements by CDC (Hall et al. 2016). Institutional Review Board approval was not required for the use of secondary de-identified data used in this study.
Data source
NVDRS is a state-based active surveillance system for violent deaths including homicides, suicides, and unintentional firearm deaths. State-level data abstractors link data from vital records, coroner or medical examiner reports, and law enforcement reports and code detailed information using CDC-developed guidelines (National Center for Injury Prevention and Control 2018). The abstractors also prepare narrative summaries from medical/coroner and law enforcement reports with incident-level details that contain information about circumstances that principiated the violent death gathered through interviews with the victim’s friends and family, suicide notes, toxicology reports, and other available information. This unique attribute of data linkage and relatively more complete information on the victim-offender relationship and homicide circumstances is a strength of NVDRS over other alternatives such as the Supplementary Homicide Reports (SHR) by the Federal Bureau of Investigation (FBI). The scope of underreporting and missing data on homicides is also smaller in NVDRS compared to SHR (Shields and Ward 2008).
For certain variables, like perpetrator information, the amount of missingness was not negligible in our NVDRS data. We have indicated the proportion of missing data for each variable in the table footnotes so that readers can exercise caution interpreting our results for certain variables that have high magnitude of missingness. We also examined whether the extent of missingness was differential based on the main categories of interest (i.e., firearm vs. non-firearm). Supplemental Table S3 indicate no notable differences between data missingness in firearm and non-firearm homicide groups.
There were no missing data for binary variables (“yes” or “no”) on incident characteristics. Nonetheless, according to NVDRS documentations (National Center for Injury Prevention and Control 2018), the response “no” in case of incident characteristic in particular could either mean absence of the circumstance or lack of confirmation of its presence. We had access to narrative summaries from medical/coroner reports and the law enforcement reports for each homicide case. For some cases, either medical/coroner reports or law enforcement or both narratives were missing. Some narratives were more detailed than the others depending on the information available to the abstractor from the medical/coroner and law enforcement reports at the time of abstraction. Both the medical/coroner and law enforcement narrative summaries included information about the context of the homicide when the information was known.
We used abstractor coded data for all older adult homicides from 2003 to 2017 for 36 available states (Alaska, Arizona, California, Colorado, Connecticut, Delaware, Georgia, Hawaii, Illinois, Indiana, Iowa, Kansas, Kentucky, Maine, Maryland, Massachusetts, Michigan, Minnesota, Nevada, New Hampshire, New Jersey, New Mexico, New York, North Carolina, Ohio, Oklahoma, Oregon, Pennsylvania, Rhode Island, South Carolina, Utah, Vermont, Virginia, Washington, West Virginia, and Wisconsin), Puerto Rico, and District of Columbia. Statewide data availability varied by year as states were progressively added into the system. To assess the representativeness of our data for the total US population, we compared the victim characteristics from our study with those from WISQARS since WISQARS has data available for all 50 states and the District of Columbia. The results were fairly comparable between the two data sources in terms of the distribution of selected characteristics (Supplemental Table S1). Fatal Injury Data from WISQARS is a source of counts and age-adjusted rates of injury mortality by mechanism and manner of death based on the International Classification of Diseases - 10th revision.
We reviewed all of the qualitative narrative summaries available for a random sample of 150 older adult homicide cases to better understand the context of older adult homicides perpetrated using firearms. To obtain the sample, we first excluded the cases where both medical/coroner and law enforcement narrative summaries were missing. We generated random numbers such that each case had a random number associated with them in the data. With ascending sorting, we kept the first 150 smallest random numbers and thus obtained 150 random cases with narrative data. In the resulting sample, law enforcement narrative summary was missing for 10 cases and medical/coroner narrative summary was missing for 2 cases. For the remaining 138 cases, both types of narrative summaries were available.
Measures
We first described homicide mechanisms based on mechanisms used for homicide including firearms, sharp instruments, blunt instruments, hanging/strangulation/suffocation, personal weapon (e.g., hands, feet, fists) or other (e.g., poisoning, fall, fire/burn, electrocution, explosives, non-powder guns, nail gun, taser, shaking, motor vehicle and other transport vehicles, intentional neglect, biological weapon, exposure to weather conditions). To compare victim, perpetrator, and incident characteristics by mechanisms, we dichotomized the injury mechanism into firearm vs. non-firearm. We reported perpetrator information for the person identified in NVDRS as the primary suspect in the incident. We dichotomized victim’s education level as obtaining less than a high school degree vs. a high school degree or above. We categorized race and ethnicity as White, non-Hispanic, Black, non-Hispanic, Hispanic, and all other races (Adhia et al. 2019). Victim-perpetrator relationship was coded as intimate partners, family members (other than the spouse), acquaintances (e.g., friends, colleagues), and strangers. NVDRS defines intimate partner as a current or ex-partner including boyfriend, girlfriend, dating partner, sexual partner, or spouse. We categorized location of injury as victim’s home, other homes/apartments, street/sidewalk or alley, parking lot/garage or motor vehicle, and other locations. When at least one additional death occurred (either a suicide or homicide) as part of the incident, we classified it as a multiple-victim incident.
NVDRS also contains several incident circumstances (e.g., homicide-suicide, intimate partner violence-related, mercy killing, justifiable self-defense, alcohol use, drug involvement, gang-related) coded as yes/present or no/not present/unknown. Complete definitions for all incident circumstances variables used in this study are provided in the appendix (Supplemental Table S2).
Statistical analysis
To investigate firearm vs. non-firearm homicides, we excluded 251 (3.9%) of the initial 6439 older adult homicides where the mechanism was unknown. A total of 5961 homicide incidents with 6188 older adult victims were included in this analysis. Homicide incident and victim numbers were different as some of the incidents included more than one victim. Age-adjusted rates were generated for older adult homicides overall and separately for firearms and non-firearm incidents using WISQARS. We used descriptive statistics to characterize firearm and non-firearm homicides in terms of victim, perpetrator, and incident features. We calculated prevalence differences and constructed their corresponding 95% confidence intervals using the exact method based on binomial distribution. STATA version 12 (StataCorp 2011) was used for all analyses.
To understand the contexts of firearm homicides among older adults using narrative data, we began with a list of common contexts identified from the circumstance variables in the quantitative analysis and from prior literature (Rogers and Storey 2019). This list expanded as themes emerged through iterative review of the records. Authors RAS and ARR examined and categorized the narratives and reached a consensus on predominant contexts. This process yielded four major themes. Themes with a smaller number of cases were grouped into the “other” category. We present results with examples created by combining information from multiple narratives with some alterations to avoid identification of victim/perpetrator identity per CDC recommendations.