Introduction

National Vital Statistics System (NVSS) data grossly miscount unintentional firearm injuries, undercounting cases among children and overcounting cases among adults (Barber, 2002; Barber and Hemenway 2011). One reason for these inaccuracies is that when assigning manner of death (MOD) for inclusion on death certificates, coroners and medical examiners do not always apply uniform criteria. (Centers for Disease Control and Prevention 2022; Hanzlick et al. 2015).

An alternative source of data on violent deaths, the National Violent Death Reporting System (NVDRS), takes steps to reduce variability in MOD coding by providing uniform criteria for assigning an “abstractor manner of death” (AMD) (Centers for Disease Control and Prevention 2022). Rules instruct abstractors, for example, to code fatalities from Russian roulette as suicides, regardless of MOD assigned by coroners and medical examiners.

A previous study focusing on early years in the NVDRS, 2003–2006, found that AMD coding for unintentional firearm deaths was accurate (Barber and Hemenway 2011), but a more recent study, published in this journal, reported that the NVDRS greatly undercounted self- and other-inflicted unintentional firearm deaths in which both the victim and (for other-inflicted injuries) the shooter were under age 15 (Vaishnav et al. 2023). In the latter study, the authors reported that between 2009 and 2018 the NVDRS correctly coded only 61% of the unintentional firearm deaths identified by the researchers (Vaishnav et al. 2023). The reason for this discrepancy and the types of cases miscoded is not clear. One possibility we attempted to assess is that the AMD is inaccurate. The current study adopts Vaishnav et al.’s case definition (and study period) to assess the accuracy of AMD relative to the manner determined by our research team after reviewing NVDRS narratives, 2009–2018. We also characterize discordant cases.

Methods

We identified all firearm deaths that occurred among the 39 states that participated in NVDRS at any point between 2009 and 2018 in which victim age was under 15 years of age, and if other-inflicted, the suspect’s age was determined to be under age 15. After reviewing incident narratives, we assigned a researcher-adjudicated MOD (RMD). RMD was compared with AMD and with the manner recorded on the death certificate. Specificity, sensitivity, and predictive values positive (PVP) and negative (PVN) were calculated for the AMD and for death certificate manner based on RMD as the gold standard.

Results

We identified 924 firearm deaths in NVDRS of which 273 were adjudicated by our team as unintentional. Comparing AMD and RMD, the sensitivity, specificity, PVP and PVN of the AMD for unintentional firearm deaths were, respectively, 90%, 99%, 98% and 96% (Table 1). 86% (24/28) of false negatives were coded by abstractors as homicides. At least 25% of the false negatives pertained to cases where the narrative explicitly noted adult negligence (e.g., negligent manslaughter) while also describing an unintentional shooting by a child, or described an unintentional shooting by a child being prosecuted for manslaughter or homicide (not shown). Comparing RMD and death certificate MOD, the sensitivity, specificity, PVP and PVN were, respectively, 63%, 99.9%, 99%, and 86%. By contrast, death certificate manner had relatively poor sensitivity (of 63%).

Table 1 Accuracy of manner of death for unintentional firearm deaths to children by children, NVDRS, 2009–2018
Full size table

Discussion

In our sample of 924 deaths, the AMD had high levels of agreement with our researcher-adjudicated manner of death. However, NVDRS abstractors were not perfect, missing 10% of cases our research team identified as unintentional firearm deaths, mostly because abstractors coded unintentional firearm deaths as homicides. While AMD coding in NVDRS is good, it could be improved if NVDRS coding guidelines explicitly affirmed that AMD should be coded as unintentional when the shooter did not intend harm and was not using the firearm to threaten or coerce another person. This includes cases where the shooter or the adult owner of the gun might be held criminally liable for negligent manslaughter, as might be the case if an adolescent unintentionally shoots another during horseplay with a parent’s gun that they assumed was unloaded. This recommendation could be applied to victims of all ages as the logic underlying it directs the abstracter to focus on the shooter’s intent, not on legal definitions of liability.