Homicide Deaths Among Adult Cohabitants of Handgun Owners in California, 2004 to 2016: A Cohort StudyFREE

We drew the cohort for this study from a parent cohort assembled for the LongSHOT project. The parent cohort consisted of 28 642 659 adult residents of California—essentially all registered voters in the state—who were followed from 18 October 2004 through 31 December 2016.

To create the study sample, we applied several exclusion criteria to the parent cohort (Figure 1). First, members of 1-adult households were excluded (they had no observable cohabitants), as were members of households with 5 or more adults (n = 6 052 439). Second, we excluded people who had acquired handguns between 1985 and the day they were otherwise eligible to enter the study cohort, unless their stock had been fully divested at that moment; their cohabitants were also excluded (n = 2 775 602). Third, we excluded people who were residing with handgun owners on the first day they appeared in the cohort (n = 401 721). Thus, all members of the study cohort began contributing observation time while unexposed (22). Some cohort members subsequently became exposed: when a person they lived with acquired a handgun, when they moved in with a handgun owner, or when a handgun owner moved in with them. Fourth, to reduce exposure misclassification, we took advantage of historical information on all lawful handgun acquirers statewide (that is, not only those present in the parent cohort) and excluded people residing with them (n = 1 170 533). Finally, people who were observed only before their 21st birthday were dropped (n = 514 298), as were people with missing birth dates or residential locations (n = 158 970). Additional details and rationales for this set of exclusions are provided in section III of the Supplement.

The parent cohort is described in detail elsewhere (23, 24). Briefly, we formed it by linking information on handgun transactions and all-cause mortality among adults in California to a series of historical extracts of the California Statewide Voter Registration Database (SVRD).

The SVRD enumerates all registered voters in the state. Because the SVRD must be kept up to date—with new registrations, deregistrations, and changes of names and addresses—extracts present a sample of adults known to be alive and residing in California on the extract date. We obtained 13 historical extracts of the SVRD. The extracts were spaced approximately 1 year apart, spanned our study period (18 October 2004 through 31 December 2016), and included approximately 61% of all adult residents of the state (Supplement Table 2).

Virtually all lawful transfers of firearms in California must be transacted through a licensed firearms dealer (25). The California Department of Justice has archived details of handgun transfers and transferees in the Dealer Record of Sale (DROS) database for decades; details of long gun (rifle and shotgun) transfers were not routinely archived until 1 January 2014 (26). We obtained records of the 9.1 million handgun and long gun transfers recorded in the DROS database over a 32-year period (1 January 1985 through 31 December 2016).

The California Death Statistical Master Files are the state's official mortality records. They contain detailed information on deaths of state residents, wherever the deaths occur. We obtained data on all deaths during the study period.

Within the parent cohort, we formed households by matching “cohabitants”—members of the LongSHOT cohort who resided at the same address during the same period—using information on primary residence provided in the collection of SVRD extracts. Section II of the Supplement describes this household matching method.

Our outcome of interest was homicide coded according to the International Classification of Diseases, 10th Revision, which classifies homicides by method (X85-X99, Y00-Y09), including firearm (X93-X95). Locations of the homicides came from the state mortality records, supplemented with information obtained in searches of the LexisNexis news database and the internet; information on the relationship of victims to perpetrators came exclusively from news searches. Details of the homicide coding methods are provided in section IV(b) of the Supplement.

The DROS data indicated which cohort members acquired handguns and dates of acquisition. Age, sex, and current residential address came from the SVRD. Racial and ethnic group and missing values for sex were imputed using validated methods (27, 28). We geocoded the residential addresses and assigned them to census tracts—geographically contiguous areas designed to approximate small neighborhoods (29).

Using DROS data, we constructed 2 additional variables. First, we identified “divestments”—transfers of the last known handgun a cohort member owned. Second, we flagged cohort members who acquired long guns with an indicator variable that switched on at the date of their first-known long gun acquisition.

For additional details on all study variables, including missing data and imputation methods, see sections IV, V, and VI of the Supplement.

The final analytic data set was at the person–period level. Eligible cohabitants entered the cohort on the date of the SVRD extract in which they first appeared. Cohort members' observation time ended on the earliest of the following: the day before the date of the next extract in which they did not appear or appeared in an ineligibly sized household (1 or >4 adults), the day before they acquired a handgun, their death or the death of a handgun owner cohabitant, or the end of the study period (Supplement Table 14).

Exposure time began on the day someone residing with a cohort member acquired the household's first documented handgun, or on the date of the first SVRD extract showing a cohort member residing with a handgun owner. Exposure time ended on the day before the date of an SVRD extract that showed the cohort member no longer residing with a handgun owner (because of household moves), the day the last handgun owner in the cohort member's household fully divested, or when observation time ended.

The study was supported by a grant from the National Collaborative on Gun Violence Research. The Fund for a Safer Future, and the Joyce Foundation and internal funds from Stanford Law School and the Stanford University School of Medicine supported assembly of the parent cohort. The funders had no role in the study's design, conduct, or reporting.

The study cohort comprised 17 569 096 adults who were followed for an average of 6.9 years. None of them had recorded handgun acquisitions between 1985 and the time they entered the study cohort, but 595 448 (3.4%) started residing with handgun owners during the study period; two thirds of these cohabitants of handgun owners were female (Table 1). Compared with cohabitants of nonowners, cohabitants of owners were more likely to be female (66.5% vs. 52.3%), be White (72.6% vs. 61.3%), and reside outside an urban area (14.5% vs. 9.1%).

A total of 737 012 cohort members died during the study period; 2293 died by homicide, of which 1495 were homicides by firearm. Fifty-three percent of the homicides occurred away from the victim's home, 37.8% occurred at the victim's home, 1.3% involved victims residing in irregular dwellings (for example, homeless or institutionalized), and the location could not be determined for the remaining 7.5% (Figure 2). Among homicides that occurred at home, the relationship of perpetrator to victim was unknown for 26.6%; among the rest, the victim was killed by a spouse or intimate partner in 36.9%, another family member in 25.9%, a friend or acquaintance in 20.9%, and a stranger in 16.2%.

Firearms were used in 65.2% of all homicides, 51.6% of homicides at home, and 77.2% of homicides away from home (Table 2). Crude rates of all 3 of these categories of homicide were higher among cohabitants of handgun owners than among cohabitants of nonowners.

In adjusted analyses, cohabitants of handgun owners had virtually the same all-cause mortality rate as cohabitants of nonowners (adjusted hazard ratio, 1.01 [95% CI, 0.99 to 1.03]) but more than double the homicide rate (adjusted hazard ratio, 2.33 [CI, 1.78 to 3.05]) (Figure 3). This higher homicide rate was primarily attributable to a higher rate of homicide by firearm (adjusted hazard ratio, 2.83 [CI, 2.05 to 3.91]). (Complete estimates from all models are provided in Supplement Tables 22 to 28.)

In analyses stratified by location of death, cohabitants of handgun owners had substantially higher rates of homicide at home (adjusted hazard ratio, 3.02 [CI, 2.12 to 4.30]). Again, this elevated risk was primarily attributable to homicides by firearm (adjusted hazard ratio, 4.44 [CI, 2.84 to 6.93]). Rates of homicide occurring away from home were also higher.

Among homicides occurring at home, the hazard ratio comparing the rate of homicides perpetrated by strangers among owners with the corresponding rate among nonowners was 1.47 (CI, 0.34 to 6.41). The hazard ratio comparing rates of homicide perpetrated by spouses and intimate partners between the 2 groups was 4.25 (CI, 2.63 to 6.86), and for homicides perpetrated by family members, friends, and acquaintances it was 2.45 (CI, 1.44 to 4.15). Rates of homicide by firearm at the hands of spouses and intimate partners were sevenfold higher among cohabitants of handgun owners (hazard ratio, 7.16 [CI, 4.04 to 12.69]), and 84% (197 of 235) of cohort members killed this way were women.

Cohort members who were living with handgun owners had higher rates of firearm homicide throughout follow-up than cohort members who were not living with handgun owners (Figure 4). The adjusted cumulative incidence of firearm homicide at 5 years was 0.012% (CI, 0.008% to 0.018%) among exposed cohort members and 0.008% (CI, 0.008% to 0.009%) among unexposed cohort members. These estimates imply that, for every 100 000 nonowners whose cohabitant acquired a handgun, 4.03 more died by firearm homicide in the ensuing 5 years than died among nonowners whose household remained gun-free.

The bias analyses showed that a putative confounder would need to be large to nullify our main results (Supplement Table 19 and Supplement Figure 5). For example, for a confounder to nullify the positive association detected between cohabiting with a handgun owner and homicide by firearm at home, it would need to both increase the risk for homicide by firearm by a factor of 8.35 and be 8.35 times more common among cohabitants of owners than cohabitants of nonowners.

The adjusted hazard ratios for overall homicide risk from analyses that excluded the first 12 months and 24 months of exposure time were 2.30 (CI, 1.76 to 3.01) and 1.90 (CI, 1.25 to 2.90), respectively. Like other key estimates (for example, adjusted hazard ratios for at-home homicide and at-home homicide perpetrated by strangers), they were not particularly sensitive to imposition of these exclusion periods (Supplement Table 18).

Estimates of homicide risk that relied on multiple imputation methods to account for missing values for sex, racial and ethnic group, and type of homicide were very similar to those reported in Figure 3 (Supplement Tables 20 and 21).