BRANAS ET AL. RESPOND

Abstract

We are grateful for Wintemute's comment and the opportunity to address the areas of our research about which he was unclear. We designed and executed a case–control study that was different than the study suggested by Wintemute, but that does not mean it was erroneous.1 Our study did indeed control for both time and place and included an appropriate, population-based control group that improved on the shortcomings of prior, related case–control work.1,2 Assaults not involving gun injury have the potential to go undetected by police and hospitals. Our case detection system likely missed few, if any, shootings in Philadelphia, Pennsylvania, over the study period. Wintemute is correct that the association we observed applies to either an increased risk of assault or an increased risk of being shot given an assault. This raises an important, but different, etiologic question to subsequently pursue, given the overall association between gun possession and gun assault that we observed. Because appropriate assumptions regarding the etiologically relevant timing between short-lived exposures and acute outcomes are important to avoid biases,3 we controlled for time through a common risk set sampling approach.1,3–10 We also controlled for place with 7 confounders including being indoors versus outdoors and numerous neighborhood characteristics. We did not “unfairly” assume that violence was randomly distributed across Philadelphia, but rather that Philadelphians were not somehow immune from being shot based on their location and that their risk changed depending on factors for which we statistically controlled. Stray bullets were just 1 argument in support of this; we also argued that guns were mobile and could be carried into practically any neighborhood street, home, or workplace environment in Philadelphia. Ours was not a study of pedestrian injury where being indoors would have essentially eliminated the risk of being hit by a car.3 Restricting our analyses to cases in which victim and shooter were thought to have some prior involvement produced an odds ratio (OR) of 9.30 (P = .03) under the same full model specification. Cases in which victim and shooter were thought to have no prior involvement produced an OR of 4.28 (P = .40). This remains consistent with our conservative interpretation that, on average, urban gun possession was not protective. Built on prior work, our study has contributed to understanding the link between gun possession and gun assault. Future studies employing similar, and alternative, designs are certainly in order to address new research questions which follow from our work.