Background

In 2008 over two million people in the USA were injured as a result of interpersonal violence.1 As a result there were 19 200 deaths; this mortality rate is more than 13 times the number of American ‘War on Terror’ casualties that same year.2 ,3 While medical and criminal perspectives tend to dominate the public discourse on violent injury, incidents have social, economic, political and personal repercussions.4 For example, the total cost of violence in the USA is estimated to be nearly 10% of the gross domestic product.5 ,6 Such figures fail to include the extensive psychological and emotional impacts of victimisation.7 This multifaceted burden of violent injury presents a significant challenge to public health policymakers and practitioners. To contribute to injury prevention and control efforts, we employ a spatial epidemiology approach that emphasises the role of urban geographies in this multivariate temporal analysis of violent injury.

Within cities, violent injury incidence has been shown to vary between neighbourhoods. For example, Brantingham and Brantingham8 point to dense, elevated violence rates along the edges of entertainment districts, contrasting low rates in larger, nearby residential areas. In order to identify high-risk urban areas (hotspots), many researchers use geographic information systems (GIS).9–11 GIS technologies and methods constitute a powerful toolset for analysing spatial data, and have been pivotal in identifying the links between violent injury and features of social and physical environments.9 ,12 ,13

Research has demonstrated that injury is strongly correlated in space-time, and is related to features of the social and physical urban environments.12 ,14–16 These relationships can be further refined by including a temporal variable. Cusimano et al identified peaks in violent injury incidence at night and early morning, tending to cluster around alcohol-serving establishments, low-income housing and homeless shelters.9 ,10 ,17–21 This work differs in that we expand our analysis to include victim characteristics, the weapon of injury and a measure of social deprivation.

Social structure and deprivation are shown to be correlated with violent injury.16 Socioeconomic indicators and indices (eg, income inequality and employment status) tend to be strong predictors of injury risk.22 ,23 We examine this relationship using a deprivation index developed by Bell et al.24

In this work, we focus on severe trauma as the result of interpersonal violence in Vancouver, Canada. Specifically, we investigate how these incidents cluster in space-time, and examine how these patterns are reflected in the nature of the victim, the injury, the physical environment and the socioeconomic deprivation level of the neighbourhood where the incident took place.

Data and methods

Injury data were acquired from the British Columbia Trauma Registry comprising the location, date, time and cause of every injury that was treated in a trauma centre in the Metro Vancouver area between 2001 and 2008, inclusive. Owing to data recording protocols, confidentiality regulations, and to eliminate minor incidents, only cases with an injury severity score (ISS)25 greater than 15 were included. As such, only severe injuries and injuries to multiple body parts were included; sports injuries and self-inflicted injuries were excluded. Cases where the patient had expired prior to arrival at a trauma centre were not recorded in the trauma registry and are therefore excluded from this analysis.

The data were classified into seven injury mechanisms: gunshot wound; sharp object trauma; blunt object trauma; bodily force (eg, punch); and other assault. A total of 575 incidents were mapped, each represented as a point in space and time. In order to determine whether an incident took place during the day or night, civil dawn and civil dusk were calculated for every day in the data period.

The following variables were used in this analysis: injury location (mapped); victim age; victim sex; date and time of incident; and mechanism of injury. Pearson's χ² test was used to determine the independence of variables against day/night, and the data were graphed in relation to time of day, day-night contrast, time of year, and day of the week. These relationships were mapped in order to identify multivariate clusters (eg, night time gunshot injuries).

The data were imported into geographic information systems and kernel density estimation was used to produce hotspot maps to contrast hotspots for men and women. As these data constitute a population, kernel density estimation was selected as a non-inferential interpolation method for visualising the spatial distribution of cases. Natural breaks (Jenks’ algorithm) was selected to identify areas of high concentration, and the cut-off was manually adjusted to obscure several cases’ locations. To identify spatial-temporal hotspots, the spatial scan statistic with a space-time permutation model was used.26–28 In the interest of patient confidentiality, the kernel density bandwidth parameter and spatial scan window size are withheld to prevent reverse calculation.

In order to explore the relationship between violent injury hotspots and socioeconomic deprivation, we used the Vancouver Area Neighbourhood Deprivation Index (VANDIX), constructed by Bell et al.24 VANDIX is a single quantitative metric constructed from numerous census variables in consultation with regional health specialists. The deprivation score for every census dissemination area was calculated, and classified into quintiles and the violent injury locations were queried in GIS to identify the proportion within each deprivation quintile.

Findings

The violent injury data comprise 518 male and 57 female victims. This disparity corresponds with findings in the literature.14 However, trauma data are certainly skewed due to the under-reporting of domestic violence.29 When categorised by day/night, we found that 75% of injuries among men occurred at night, compared with 56% of injuries among women; the association between victim sex and day/night was significant (χ² p=0.002). Figure 1 displays the temporal distribution.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Temporal distribution of violent injuries through 24 h period. Seventy-three per cent of incidents took place at night. Note the bimodal peaks at midnight and 2:00. These peaks clustered spatially within the nightclub district (23:00–3:00) and the Downtown Eastside (00:00–4:00). The peak between 2:00 and 3:00 coincides with required bar closing times.

One in every three injuries in this dataset was caused by a firearm. However, weapon of choice and time/location do not significantly correlate, matching findings in the literature.30 Figure 2 divides each mechanism of injury into day/night incidence. The difference of day/night ratios suggests that the severity of a mechanism is related to its proportional daytime prevalence. That is, an offender may feel more secure using a firearm under the cover of darkness, whereas the distinction is less consequential if throwing a punch. Note that because these data only include severe trauma cases, blunt object and bodily force injuries are under-represented.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Day/night comparison of violent injury counts by mechanism of injury. All injuries are more prevalent at night while firearms and injuries from sharp objects (eg, knives) are much more prevalent. The increase between day and night incidence of firearm injury is 380%, compared with 247% for sharp object and only 183% for bodily force.

In figure 3, incidents are categorised by day of the week. Each night is defined as continuous until the following dawn; that is, Saturday night continues until sunrise on Sunday. This is done in order to account for alcohol consumption initiated in the evening and prolonged through the night. This weekend spike is to be expected, but note the ‘hangover effect’: relative peace on Saturday and Sunday during daylight hours.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Day/night contrast of violent injury counts by day of week. There is a clear spike on Friday and Saturday nights with daytime figures remaining relatively constant throughout the week. Friday night and Saturday night account for one-third of all injuries. While the median age for all injuries in the dataset is 30, it drops to 25 on Friday and Saturday nights. This spike clusters around the nightclub district.

Figure 4, below, displays violent injury hotspots in the Metro Vancouver area. This pattern is similar to the distribution of low-income, high-density housing in the region. The most prominent hotspot (bright red) occurs in the Downtown Eastside, and some minor hotspots are found in commercial zones of the outer suburbs. The linear trend from Downtown to New Westminster features a series of minor hotspots which occur close to Skytrain stations (light rail transit) and a major shopping centre.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Injury hotspots by subcategory. Note that the Downtown Eastside is an injury hotspot regardless of the breakdown of victims. However, male victims of violent injury are spread along other corridors (eg, Commercial Drive) as are night-time injuries. Note also the clustering of cases in the Downtown Eastside, a neighbourhood in the top quintile of socioeconomic deprivation.

Also evident are spatial differences in female (red) and male (blue) victim hotspots. While the sample size for women is comparatively small, when mapped, some geographic differences are notable. The club district, Commercial Drive and the stadium feature only male victims, while victims in Chinatown are all women.

The circles represent cluster locations detected using the space-time permutation model of the spatial scan statistic, with the time window in which the cluster occurs. The nightclub district, the Downtown Eastside and Commercial Drive contain night-time hotspots that are particularly active on Friday and Saturday nights. These neighbourhoods have a high concentration of alcohol-serving establishments that feature live music and late closing times. The stadium and arena also appears as a hotspot. That this occurs during the night suggests that these cases may also be attributed to the high concentration of alcohol-serving establishments surrounding, and including, the stadium.

When violent injury locations were compared with the VANDIX24 we found that 77% of all severe violent injuries occurred within neighbourhoods in these two most deprived quintiles. There are two notable exceptions: the club district and Metrotown. While the cluster in the club district is explained by the high night-time incidence in proximity to alcohol-serving establishments, the case in Metrotown is more complex and better explained by features of the built urban environment.

Implications for prevention and control

To examine these findings within an injury prevention and control context, we consider potential interventions actions within education, regulation, the built environment and injury surveillance, recommending a comprehensive comparative study of their effectiveness in other settings.

Physicians occupy a valuable position to discuss violence with patients and refer them to community and health resources.43 A 1995 study, Hedges et al44 found that nearly half of all repeat trauma patients were injured by the same mechanism that caused their first visit. As such, point-of-care interventions may prove useful in behaviour modification. However, upstream interventions hold strong potential; in this case, information posters could be placed in nightclub washrooms, a technique used in antismoking campaigns.45 However, offender-targeted campaigns have been shown in the literature to have minimal effect on offender opinions or behaviours, and the management at alcohol-serving establishments may be reluctant to engage in campaigns that may discourage consumption.46 ,47 Conversely, the WHO recommends campaigns targeting the cultural norms that promote violence, with an emphasis on youth education.48 Such programmes would require broader policy support and an acknowledgement of the ‘lag effect’ of youth education on adult health outcomes.

The most prominent temporal spike in violent injury found in this analysis data points towards the ‘closing-time’ debate. There is strong evidence that ‘last call’ policy can reduce excessive consumption and resulting harms,1 ,48 ,49 although the Licensing Act 2003, which allowed for longer serving hours in England and Wales, appeared to have no significant effect.50 Standardised closing times may contribute to increasingly rapid drinking as last call approaches, leading to higher peak levels of intoxication. One policy alternative is the ‘staggered closing’ strategy, where venues may purchase liquor licenses priced along a range of closing times. This is intended to prevent high concentrations of intoxicated patrons at closing time. The WHO recommends strategic pricing to reduce alcohol-related harms, which has been shown both in England51 and British Columbia52 to be an effective policy tool for reducing negative health outcomes.48

There were 192 gunshot trauma cases in this dataset, more than any other mechanism. However, the exclusion of minor injuries in our dataset inflates the proportion of severe trauma, limiting any inference. The WHO global report on violence prevention points towards a widespread weapons availability, pointing towards background checks, safe storage regulations and more restrictive regulation.48 While Canada has relatively strong firearms legislation, the majority of illegal firearms in the Metro Vancouver region are smuggled from the USA or overseas.53 Improved border checks are therefore a potential avenue for intervention.

While increased police presence in hotspots is often recommended,31 ,32 ,54 Goldstein proposes problem-oriented policing as a method that emphasises the role of the police officer to identify underlying drivers of crime, build relationships with the citizenry and acknowledge the limits of the criminal justice system to address problems.55 This approach has been tested in the field with promising results,29 ,31 including ‘community policing’ in Vancouver.

Planning and modification of the built environment provide another direction through which violent trauma can be prevented.31 Stevenson describes urban violence with the enclosure/encounter models.15 A crowded bar is an example of a built enclosure, while the crowded street at closing time causes numerous encounters, both of which can contribute to interpersonal violence. Modifications could include improved nightclub design and wider sidewalks.

Effective injury surveillance programmes can facilitate goal-tracking43; social web technologies are a promising asset in this regard, particularly in deprived urban areas where mobile phone usage is rapidly growing.56 High-quality trauma data were vital in this and other studies. Research and subsequent interventions would benefit from the implementation of a broad violence surveillance programme that combines data from police, medical, schools, community resources (such as women's shelters), legal records and other sources.

In this study, we found several distinct patterns of violent injury with varying characteristics between neighbourhoods. As such, surveillance and intervention programmes should be custom-designed for individual neighbourhoods.31 ,55 While education, regulation, the built environment and surveillance programmes hold strong potential to reduce violent injury in urban space, successes based on policing and accessible medical services have been quantitatively demonstrated.31 ,54 With the ultimate aim of injury control, a successful campaign will encompass a range of strategies to confront the issue upstream (eg, liquor policy and education) and downstream (eg, policing and ample ambulatory resources). This study is limited by the exclusion of cases with a severity score below 15; as a result, these findings are applicable only to severe trauma. Further, the inclusion of patients who expired prior to arrival at a trauma centre may have an impact on the distribution of hotspots. Their exclusion may have led to an under-representation of gang-related violence, which caused numerous deaths during the study period. Future spatial analyses of violent trauma may benefit from the integration of multiple data sources to include a broader range of severity and an extended temporal period.

In this paper, we identified spatial–temporal patterns of violent trauma in Vancouver, Canada, described the neighbourhood-specific contexts within which the examined hotspots are found, and discussed some potential intervention strategies. While this analysis was limited to severe trauma in one city, the findings and implications may prove useful for enriching research efforts in other urban places. However, further research should examine ways to refine the GIS-based methods used herein, conduct more place-specific contextual investigation and examine the efficacy of intervention strategies in other study areas.