Sample Selection
The sample identification protocols for this study have been previously published (17). The cohort was recruited between Fall 2013 and Spring 2014 in south central Pennsylvania, USA. Women were eligible for inclusion if they screened positive for a lifetime history of IPV based on the humiliation-afraid-rape-kick (HARK) screening instrument, a validated 4-item screen to identify IPV in healthcare settings (18), and left their contact information after completing the screening questionnaire. Data collection for the one year follow up took place between Fall 2014 and Spring 2015.
Briefly, we identified a sample of 24,338 women ages 18-64 in south central Pennsylvania with least one primary care visit in the past year. A randomly selected subset of 2,500 women were invited to participate; surveys were received from 1,191 women from the clinical sample. The sample was stratified for rural residence using the zip-code based approximation of the Rural-Urban Commuting Area codes, a classification system based on city size and commuting practices (19). Rural-residing women were oversampled to achieve appropriate numbers for analysis. To augment the cohort drawn from the healthcare setting, posters were also displayed at 26 domestic violence shelters in Central Pennsylvania, inviting women to participate in the survey online, by phone, or by mail. From this population, an additional 73 women were recruited to participate in response to these posters, yielding the final sample size of 1264 women who completed the screening survey.
Among this sample, those women who screened positive for lifetime exposure to IPV based on the humiliation-afraid-rape-kick (HARK) screening instrument, a validated 4-item screen to identify IPV in healthcare settings (18) and who left their contact information were contacted with an invitation to participate in a longitudinal study, requiring completion of a survey at baseline and one year later.
Of the women recruited via the ambulatory cohort, 500 women screened positive for IPV, and 270 participated in the baseline survey. From the participants recruited from the shelter sample, 60 women screened positive for IPV, and 38 participated in the baseline survey. These two subgroups were treated identically after initial recruitment. After one year, all women who completed the baseline survey were contacted and asked to complete the follow up survey. Among the women completing the follow up survey, 239 from the ambulatory cohort and 28 shelter participants (for a total 267 women) participated in the one-year follow-up survey. These 267 women form the analytic cohort for this analysis.
Study data were entered and managed within REDCap (Research Electronic Data Capture), a secure, web-based application designed to support data capture for research studies, hosted by Pennsylvania State University (20). This study was conducted with approval from the Institutional Review Board (IRB) for all study protocol and study documents. All women reviewed a written or verbal informed consent and consented to participate in this research. To protect participants further, and due to the sensitive nature of this study, a Certificate of Confidentiality (CC-MH-12-204) was obtained from the National Institutes of Health for this research.
Variables of Interest
The follow up cohort of 267 women was assessed for the three primary outcomes surrounding firearms, of 1) perceptions of access (“How easy is it for people who live near you to get a gun?”), 2) perceptions of safety (“Does having a gun around make you feel safer or less safe?”), and 3) firearm proximity (“Are any firearms kept in or around your home?”) (21, 22). As noted in Figure 1, we hypothesized that trauma exposures would affect perceptions of firearm safety, in that women with a history of trauma would feel less safe around guns, be less likely to have guns in the home, and perceive guns to be readily available in their community.
Our primary independent variables were demographics and trauma exposures. To assess prior history of trauma, participants were screened for IPV recency (past-year vs. lifetime) and IPV type (physical vs nonphysical) using HARK (18). Nonphysical IPV (humiliate-afraid) and physical IPV (rape-kick) were mutually exclusive categories, and participants were stratified into the physical IPV category if they had ever experienced physical IPV. The HARK question stem was modified to determine whether they had experienced IPV in their lifetimes compared to the past year. Additional interpersonal trauma exposures were unwanted sexual exposure (23) and adverse childhood experiences (24). “Unwanted sexual exposure” was categorized as never, lifetime, or past-year (23). “Adverse childhood experiences” (ACEs) were stratified by severity into tertiles. ACEs were determined using a definition taken from the ACE study (a collaborative research endeavor funded by the CDC and Kaiser Permanente (24).
To control for variation in the sample by demographics, we evaluated our cohort for age, marital status, urbanicity, poverty, education, and race/ethnicity. These variables were chosen because of their relevance to IPV, as well as their role in gun ownership trends. We considered whether our patients were near poverty (defined as 125% of the national poverty line) or not near poverty. Gun ownership also varies with region of the country, but our cohort is from within the same regional area, so we were unable to account for this variation.
Data Analysis
All variables were summarized with frequencies and percentages. Binomial or ordinal logistic regression, depending on the format of the outcome variable, was used to determine any unadjusted bivariate associations between each of the demographic and trauma exposure variables and each of the three firearm perception questions. Covariates were selected and retained for inclusion in the model based on their relationship to the outcomes variables as seen in the literature. As there were very few missing data, these were not included in analyses. We did not infer any missing data.
As noted above, significant data exists on the demographic variables associated with gun ownership, especially surrounding age, race/ethnicity, rurality, marital status, education, and income. Given that our outcomes variables included questions of guns in the home and also gun perceptions, we considered that these demographic variables were likely predictors of our outcome variables to be included in our analyses. To assess the relationship of gun ownership and perceptions with types of interpersonal trauma, we looked at different types of IPV, recency of IPV, unwanted sexual exposure, and ACEs to evaluate if these traumas were related to our outcomes variables. Interactions were not specifically tested in this model.
Multivariable analyses examined the associations of these exposure variables collectively with each of the three firearm perception questions while controlling for the demographic variables. All of the independent variables were tested for multicollinearity prior to inclusion in the model using variance inflation factor (VIF) statistics, and the fit of the multivariable models was assessed using the Pearson, Deviance, and Hosmer and Lemeshow goodness-of-fit tests. If the majority of these tests showed good model fit with p>0.05, we accepted the model as having good fit, and this was the case for all three multivariable models. All analyses used a significance level of p<0.05 and were performed using SAS version 9.4 (SAS Institute, Cary, NC).