Article Text
PDF
Abstract
Introduction Among the factors related to survival among individuals with gun shot wounds (GSW) is distance to trauma care. Relatively little is known about neighbourhood-level patterns of GSW mortality and distance to hospitals with trauma centres. This study focuses on distance to the nearest trauma centre as a correlate of survival among GSW victims.
Methods Fatal and non-fatal shooting incident data for 9,205 victimisation in Detroit, Michigan between 2011 and 2017 were collected. A Bayesian conditional autoregressive model was utilised to estimate block-group levels of GSW mortality. Clustering techniques were used to identify spatially proximate neighbourhoods with higher or lower than expected rates of GSW mortality.
Results Distance to the nearest trauma centre was associated with a 22% increase in fatal outcomes, per-mile (OR 1.22, 95% CI, 1.06 to 1.40) after adjusting for block-group level covariates. A Getis-Ord Gi* analysis identified 91 block groups with lower than expected rates of GSW mortality and 12 block-groups with higher than expected rates.
Conclusion Distance to the nearest trauma centre is associated with GSW victim survival. Clusters of block-groups with below-average GSW mortality were observed within close proximity of major trauma centres in Detroit. Improving speed and access to trauma care may play a role in reducing GSW mortality.
- firearms
- spatial analysis
- penetrating injury
- public health
Statistics from Altmetric.com
Introduction
Deaths from firearm injuries remain among the most common causes of death, especially among young, minority men.1 Among research studying correlates of firearm death, several studies have attempted to determine factors that differentiate fatal and non-fatal shootings.2 Indeed, the difference between a fatal shooting and a non-fatal shooting can reflect a number of variables: such as whether the shooting occurred indoors or outdoors, the calibre of gun and the location or severity of injury.3–5 As medical care has improved so has the ability of shooting victims to survive critical injuries.6 Therefore, access to timely and appropriate care is important among victims whose injuries may be life threatening. Examining hospital admissions for victims of gun violence is likely a useful tool because many individuals—including offenders—report going to the hospital for treatment after being shot.7 However, relatively few studies have studied whether distance to the nearest hospital or trauma centre is related to survival among shooting victims. One study in Chicago found that victims who were shot outside of a 5-mile radius of the nearest trauma centre were about 20% more likely to die of their injuries after controlling for a number of other factors.8 The results of this study prompted the coining of the term ‘trauma deserts’, which refer to neighbourhoods with limited access to emergency medical care.
A noted weakness of this study was the authors’ decision to dichotomise the distance to a trauma centre as either within or not within a 5-mile radius (which they note as ‘somewhat arbitrary’). The authors also make a limited attempt to spatially differentiate regions of the city with higher and lower rates of gunshot wound mortality. Instead, their analysis focuses primarily on individual-level estimates. This is not to diminish their important study; rather, it points to the need to understand the spatial dynamics of access to trauma care among victims of gun violence. To fill this gap, we analysed 7 years of fatal and non-fatal shooting victimisation data in the city of Detroit to address two primary questions:
Are shorter distances to the nearest trauma centre associated with lower rates of gunshot wound mortality?
Are there spatial patterns in neighbourhoods with below-average or above-average rates of gunshot wound mortality?
Methodology
We used data from the Detroit Police Department’s records management system regarding all fatal and non-fatal shooting incidents within the city from 2011 to 2017. We excluded from our analysis all cases involving intentional self-inflicted gunshot wounds, suicides and shooting victimisations where the hospital was reported as the incident location. This latter case often occurs when victims refuse to disclose the location or circumstances in which they were shot. This final dataset included information about 9205 victims that were distributed across 879 census block groups. The block group reflects the smallest level of aggregation for which census sample data are available. Because our analysis was concerned with neighbourhood-level determinants of gun violence lethality, we compared the observed number of fatal shootings to the expected number of fatal shootings. Within Detroit, the ratio of non-fatal to fatal shootings was about 5:1, with about 21.3% of shooting incidents resulting in a fatality.
For trauma centres, we used data on the location of all hospitals with a Michigan-designated trauma facility nearest to the city of Detroit. In total, we identified nine hospitals with trauma facilities that were closest to Detroit (the hospitals included: St. John Hospital & Medical Center (level I), Henry Ford Hospital (level I), Children’s Hospital of Michigan (level I paediatric), Detroit Receiving Hospital (level I), Sini-Grace Hospital (level II), Ascension Providence Hospital Southfield (level II), Beaumont Hospital Wayne (level II) and Beaumont Hospital Grosse Pointe (level III). Data on trauma centers are available at: https://www.michigan.gov/documents/mdhhs/List_of_Designated_Trauma_Facilities.5.9.17_571865_7.pdf). Distance to the nearest trauma centre was calculated by the Euclidian distance from the block group centroid. On average, the distance to the nearest trauma centre was 2.2 miles, which varied between 4.8 miles and 0.06 miles. Figure 1 displays the distance, in miles, from each block group centroid to the nearest trauma centre.
Distance, in miles, to the nearest trauma centre by census block group. Trauma centre locations are displayed as red circles.
In order to appropriately account for the spatial dependence of the data, we used a Bayesian conditional autoregressive model (CAR) using the ‘brms’ package in R.9 A CAR model accounts for correlations between areal units and shares information between nearby units.10 One particular use of CAR models is the ability to estimate values at locations that smooth over the observed data by using values from nearby locations.10 This smoothing reduces the impact of extreme rates in low occurrence regions (ie, block groups with a 100% fatality rate or a 0% fatality rate) and adjusts SEs for the correlated nature of the data. The ‘conditional’ portion of the CAR model comes from the process by which each element of the process is specified conditional on the values of the neighbouring units. This accounts for correlation between geographically proximate units.11 To model the spatial dependence of the block groups, we created a spatial weights matrix defining all first-order neighbours, which were defined as any polygon sharing a border with another polygon.12
We modelled fatal shootings at each block group as a function of distance (in miles) to the nearest hospital with a trauma centre. Model fit statistics (leave-one-out cross-validation) were used to select a quadratic term for the distance variable over a single linear term. Our outcome variable represented a discrete number of counts, so we used a Poisson regression estimating the number of fatal shooting incidents with the expected number of fatal shootings applied as an offset. The expected number of fatal shootings was calculated by multiplying the observed number of shooting victimisations by the population-level fatality rate of 21.3%, with mortality estimates reflecting the population-level average effect. Neighbourhood-level covariates representing demographic variables and indicators of poverty and disorder were included in the model, which were obtained from the American Community Survey 2015, 5-year estimates. After obtaining estimates from the CAR model, we used a Getis-Ord Gi * analysis to identify local indicators of spatial autocorrelation, specifically, spatially proximate clusters of block groups with higher or lower observed fatal shooting victimisations than expected.13 The Gi * statistic returns local estimates of spatial autocorrelation by identifying regions where high or low values cluster. Statistically significant clusters are identified by comparing the local observed sum to the expected local sum.
In order to identify clusters, we used the posterior mean estimate from the Bayesian CAR model. In line with recommendations from Getis and Ord,13 we applied a Bonferroni-type correction to our results to adjust for the number of independent analyses and reduce the possible ‘false-discovery’ rate. Adjusted p values were obtained using the critical Z-values, which were subsequently used to identify statistically significant regions of clustering at 90%, 95% and 99% levels of significance. With almost 900 block groups, our critical Z-values were 3.67, 3.85 and 4.17, respectively.
Results
Table 1 displays the results from the CAR model with the β coefficients, SEs and 80% and 95% credible intervals highlighted. Because the modelling strategy reflected a Bayesian methodology, no p values are generated, rather the focus is on the strength and variation of the effect (reflected in the 80% and 95% credible intervals). As shown in the model, distance to the nearest trauma centre was associated with an estimated 22% increase in fatal outcomes per mile after adjusting for block group level covariates. The quadratic term indicated that this effect decayed at larger distances by about 4% per mile. Based on the credible intervals, there was an 80% probability that the main effect of distance from a trauma centre could have varied between 9% and 36% and a 95% probability that it was between 6% and 40%. This relatively wide uncertainty reflects the underlying variation in GSW mortality and the small number of block groups that are very near to hospitals with trauma centres.
Bayesian CAR Poisson regression predicting gunshot wound mortality
While this model supported our hypothesis that distance to a hospital trauma centre played a role in neighbourhood-level GSW mortality, we were especially interested in using the model estimates to identify clusters of block groups with unusually high or low GSW deaths given the underlying city-wide rate. Figure 2 shows the results from the Getis-Ord Gi * analysis using the estimates from the CAR model. Below-average clusters are highlighted in blue, while above-average clusters are highlighted in red. The block groups are shaded according to their level of statistical significance (90%, 95% and 99%), adjusted for multiple comparisons via a Bonferroni adjustment.
Statistically significant clusters of below-average and above-average rates of GSW mortality. Below-average rates are shaded in blue, while above-average rates are shaded in red. Trauma centre locations are displayed as red circles. GSW, gunshot wound.
Table 2 displays descriptive statistics for the identified clusters. The largest cluster of below-average block groups lay predominately in Detroit’s downtown corridor. These block groups represented areas immediately adjacent to two large medical campuses: Henry Ford Hospital and Detroit Receiving Hospital, which are both designated level I trauma centres. Two other clusters lay adjacent to Sinai-Grace Hospital in the North-Western portion of the city and on the city’s east side adjacent to Ascension St. John Hospital. In contrast to the clusters directly adjacent to trauma centres, a large cluster of below-average block groups lay along the northern border of Detroit adjacent to 8 mile Avenue. These block groups were not immediately adjacent to any trauma centres. Another smaller cluster was identified in the south-Western portion of the city.
Descriptive statistics for observed clusters
In total, these clusters of block groups represented 91 locations with significantly lower rates of GSW mortality than expected. Within these clusters shootings were fatal, on average, 18.2% of the time, compared with the city-wide average of 21.3%. The median distance from these block groups to the nearest trauma centre was about 0.77 miles. In contrast, there were 12 block groups that comprised the above-average block groups. These block groups lay predominately in the north central portion of the city. On average, the percentage of shootings that were fatal in these block groups was about 24.2%. These block groups had a median distance of 2.8 miles to the nearest trauma centre. Proportionally, there were many fewer clusters of above-average rates, compared with the below-average locations.
Discussion and conclusion
Prior research has highlighted the importance that trauma centres play in the survival of shooting victims. Using a set of spatial analysis tools, we identified clusters of block groups with unusually higher or lower GSW morality. In contrast to the original study by Crandall and others, our strongest finding were clusters of neighbourhoods with below-average GSW mortality (trauma ‘oases’).8 GSW victims in these clusters died at a rate that was 14% lower than the city-wide average. This reflects a 3.1 percentage point difference: a small but certainty not insignificant margin, especially when there are more than 1000 shooting victimisations every year in Detroit.
While several of the below-average clusters (including the largest one in the downtown and midtown corridor) were within close proximity to trauma centres, another large cluster was not. This is an intriguing finding because it suggests that trauma centres play some role in neighbourhood-level GSW mortality outcomes and suggests that other neighbourhood-level factors may be at play. Other issues, such as emergency medical response time or transportation availability may also play some role here. Somewhat consistent with Crandall and others, we also identified a small number of above-average GSW mortality clusters. Victims shot in these regions died at a rate about 12% higher than the city-wide average (a difference of 2.9 percentage points).
The implications from this study appear clear. While many factors contribute to fatal outcomes in shooting incidents, increased access to timely trauma centre care likely plays a role in victim survival. In neighbourhoods with less access to trauma centres, victims on the cusp of life or death may be more likely to succumb to their injuries. Therefore, public health programmes seeking to reduce fatal gun injuries should identify neighbourhoods with lower levels of access to health and emergency services and increase the ability of victims to obtain critical injury care. Obtaining medical care during the critical ‘golden hour’ is especially important among GSW victims. While constructing large trauma centres is likely not feasible in many communities, increased EMS response times and on-site trauma care may play a role in reducing fatal outcomes among shooting victims.5 14–16
While this study identified spatial patterns in GSW mortality, there are likely many other important factors that contribute to survival in a shooting incident. For instance, wound location, injury severity, age, sex and suicidal ideation are noted contributing factors to survival.8 We recognise that our study cannot account for all of these factors, rather we have attempted to describe neighbourhood-level patterns of GSW mortality. Reducing gun violence requires addressing a multitude of issues, both individual level and community level. Increasing access to emergency healthcare represents only a single facet of this complex problem. We believe that further research should look at how public health agencies and emergency services can better address critical health issues among individuals living in underserved neighbourhoods. For instance, using non-medical first responders as a way to decrease time to trauma care may play a role in reducing GSW mortality.17 Along with increasing access to trauma centres, further research should investigate other neighbourhood-level factors that contribute to improved outcomes among the critically injured. Based on this research and other studies, we believe trauma centres are but one link in a chain of factors that lead to survival among GSW victims. The conclusions of our study point to the importance that medical care can play in reducing the burden of gun-related injuries and deaths. Ensuring speedy and quality medical care is a key facet in reducing GSW injuries and fatalities.
What is already known on the subject
Timely access to emergency medical care is associated with lower gunshot wound (GSW) mortality.
Morality rates are elevated among GSW victims who are injured more than 1.6 kilometers from a trauma centre.
What this study adds
We identify neighbourhood-level patterns of GSW mortality in Detroit, linking continuous distance to the nearest trauma centre with victim outcomes.
We identify clusters of below-average GSW mortality in neighbourhoods close to trauma centres, as well as a small number of above-average GSW mortality in neighbourhoods further from trauma centres.
We find evidence of ‘trauma oases’, which are contiguous regions with greater access to critical medical care.
References
Footnotes
Funding This project was supported by Award No. 2013-R2-CX-0015, awarded by the National Institute of Justice, Office of Justice Programs, US Department of Justice.
Disclaimer The opinions, findings and conclusions or recommendations expressed in this publication/program/exhibition are those of the author and do not necessarily reflect those of the Department of Justice.
Competing interests None declared.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Data may be obtained from a third party and are not publicly available.
