Background The National Violence Surveillance Network (NVSN) of emergency departments (ED), minor injuries units and walk-in-centres in England and Wales has brought clarity to contradictory violence trends derived from crime survey and police data. Gender, age-specific and regional trends in violence-related injury in England and Wales since 2010 have not been studied.
Methods Data on violence-related injury were collected from a structured sample of 151 EDs in England and Wales. ED attendance date and age and gender of patients who reported injury in violence from 1 January 2010 to 31 December 2014 were identified from attendance codes, specified at the local level. Time series statistical methods were used to detect both regional and national trends.
Results In total, 247 016 (178 709 males: 72.3%) violence-related attendances were identified. Estimated annual injury rate across England and Wales was 4.4/1000 population (95% CI 3.9 to 4.9); males 6.5/1000 (95% CI 5.6 to 7.2) and females 2.4/1000 (95% CI 2.1 to 2.6). On average, overall attendances decreased by 13.8% per year over the 5 years (95% CI −14.8 to −12.1). Attendances decreased significantly for both genders and all age groups (0–10, 11–17, 18–30, 31–50, 51+ years); declines were greatest among children and adolescents. Significant decreases in violence-related injury were found in all but two regions. Violence peaked in May and July.
Conclusions From an ED perspective, violence in England and Wales decreased substantially between 2010 and 2014, especially among children and adolescents. Violence prevention efforts should focus on regions with the highest injury rates and during the period May–July.
Statistics from Altmetric.com
Understanding levels and trends in serious violence, both nationally and regionally, is important to inform policy and to direct violence prevention strategies.1 ,2 Traditionally, violence levels and trends in England and Wales have been measured using police records and the Crime Survey for England and Wales (CSEW: formerly the British Crime Survey, BCS). The main strength of the CSEW is that it has retained the same methodology (face-to-face interviews) since its introduction in 1981 and through rigorous sampling practices, closely represents a reliable and valid population-level measure.3 According to the survey, 1.3 million people were victimised violently in 2013/2014 (634 000 with injury).4
In comparison, the police in England and Wales recorded 634 000 incidents of violence against the person offences during this time.4 Reasons for this disparity most likely reflect differences in ascertainment and recording practice. Most importantly, police records reflect the extent to which incidents are reported. Fear of reprisals, inability to identify assailants, lack of benefit for the injured and an unwillingness to have one's conduct scrutinised are reasons why many incidents are not reported to the police.5 Furthermore, changes in police recording practices, starting in April 2002 with the introduction of the National Crime Recording Standard (NCRS; a victim-focused approach), led to a 22% rise in the number of violent offences recorded by the police.6 More recently, a report by HM Inspectorate of Constabulary, which considered more than 8000 reports of crime in England and Wales, concluded that a third of all violent crimes reported to the police were not recorded.7 Consequently, in 2014, the UK Statistics Authority de-designated police recorded crime statistics as a national statistic.8
Over the past decade, the National Violence Surveillance Network (NVSN) of over 100 emergency departments (ED), minor injury units (MIUs) and walk-in centres in England and Wales, has clarified national violence trends. ED data represent a harm-based source of information on violence, namely, violence which results in injury serious enough to require medical treatment. ED data are, therefore, a more objective measure of violence trends than trends derived from national crime survey or police records and is the only comprehensive perspective of violence across the life course. Violence-related injury is an established category in most ED software packages and a new record is created for each individual attendance, while information relating to the time, date and location of the incident, as well as information on weapon use and assailants, is also recorded.9 This is now a mandatory requirement in England following the publication of the new Information Standard to Tackle Violence (ISTV).10 The importance of this injury-based measure of violence was recognised in ‘The Coalition: Our programme for government’, which included a commitment to reduce violence through the use of anonymised information collected in EDs following the 2010 general election.11
The aim of this study was to investigate trends in violence according to ED data in England and Wales, over the 5-year period, 1 January 2010–31 December 2014. This study set out to determine age, gender and region-specific rates of violence, violence trends and violence seasonality.
EDs and injury records
Attendance date, age and gender of patients who reported injury in violence were collected from a structured sample of 151 type 1, 3 and 4 EDs (type 1=consultant led 24 h service with resuscitation capabilities; type 3=other ED/minor injury unit; type 4=National Health Service walk-in-centres) across the nine English regions and in Wales between 1 January 2010 and 31 December 2014 (figure 1). All 151 EDs were certified members of NVSN and were recruited on the basis of their willingness to share anonymised electronic data and compliance with the provisions of the 1998 Data Protection Act and Caldicott guidance.12 From the injury records provided, it was not possible to identify individuals and data were collected under the auspices of the Freedom of Information Act.
Violence-related attendance data and statistical methods
ED attendance data received from the NVSN hospitals were categorised by gender and five age groups (0–10, 11–17, 18–30, 31–50 and 51+ years); the same categorisation to that reported in previous NVSN publications.13–15 Prior to 2011, England was subdivided into nine Government Office Regions reflecting administrative boundaries (ie, Eastern, East Midlands, London, North East, North West, South East, South West, West Midlands and Yorkshire and Humberside). Although no longer having officially devolved functions, Government Office Regions are still the primary classification for presenting regional statistics. Consistent with this, data were aggregated by month, age category and the region in which the ED was located (including Wales).
Since approximately two-thirds of EDs in England and Wales are not certified members of the NVSN, attendance data were weighted in order to account for unequal regional coverage using a series of annually updated, region-specific coverage ratios. The method used to calculate coverage ratios is shown below and has been subject to peer-review.13–15
Here, B is the total annual all-cause ED attendances sampled in region i and A is the total annual all-cause ED attendance for all EDs (including those sampled) in region i. The coverage ratio (CR) for region i is therefore the annual ratio of all-cause ED attendance at EDs sampled (B) in region i relative to the total all-cause attendances within all EDs (A) in region i. Thus, a CR equal to one indicates full regional coverage and a CR equal to zero indicates no regional coverage. ED activity statistics, containing provider level all-cause ED attendances for England, are freely available and were accessed from the Department of Health and NHS websites; provider level attendances were aggregated to give annual, regional totals. For Wales, monthly all-cause ED attendance figures were accessed from the Welsh Government website as part of its monthly report on ED waiting times; data at local health board level were aggregated to form annual totals.
Comparative regional violence data were obtained by multiplying the number of persons injured in violence in each region by 1/CR. As the total resident population by region and by age is known, it was possible to stratify the regions according to the proportion of persons injured in violence.16 A measure of the likelihood of being injured in violence and receiving ED treatment within a region (Vi) was given by
Where 1/CR is the coverage ratio for region i, ni is the number of injured persons attending EDs in region i, and Ni is the total population of region i. Violence injury rates (number of injured per 1000 resident population) for all 10 regions were computed for both genders and all age groups. A jackknife resampling technique was used to calculate 95% CIs for the violence-related injury rates.
Two separate regression models were used to estimate violence trends; a negative binomial regression was applied on the numerator (count of violence-related attendances) and a log-linear regression on the denominator (population figures). Violence rate and trend coefficients were then estimated from the ratio of these regression models. Since violence-related injury was measured by count of ED attendances, there were a substantial number of zeroes in the 0–10 years and 51 years and over age groups. A negative binomial regression model was fitted to account for any over-dispersion, using a time trend (month number) with age group, region, month and gender dummy variables. Interactions between region and time, age group and time, age group and region, gender and age group, month and age group, gender and region and gender and time were also identified. In total, there were 5904 observations from 1 January 2010 to 31 December 2014. The 2011 attendance data for the North East region were not available, resulting in 96 missing observations. In order to account for population change over time, a log-linear regression was carried out and modelled on person-days (an exposure variable taking account of month length) from census data across the same time period (also disaggregated by age group, gender and region).
In order to calculate violence rate and trend coefficients by a particular subgroup, numerator and denominator coefficients were calculated by substituting for the appropriate dummy variables in the regression equation. A similar weighted averaging process was also performed over the coefficient covariance matrix for the negative binomial regression, in order to estimate the SEs of the aggregated coefficients and assess the significance of the trend over time. Incorporating census data as ‘offsets’ in the negative binomial regression model produced the same results.
The statistical software package ‘R’ was used to clean the ED data and to calculate the violence-related injury rates.17 The same software was also used for model fitting and trend analysis.
Violence-related injury rates
In total, 247 016 violence-related attendances were recorded in the 151 EDs across England and Wales between 1 January 2010 and 31 December 2014. Disaggregation by gender and age group indicated that the sample was predominately male (n=178 709, 72.3%) and aged between 18 and 30 years (n=122 214, 49.5%, table 1).
Overall in England and Wales, 4.4/1000 residents (95% CI 3.9 to 4.9) attended EDs annually for treatment of violence-related injury between 2010 and 2014. Males (6.5/1000 residents; 95% CI 5.6 to 7.2) were nearly three times more likely than females (2.4/1000 residents; 95% CI 2.1 to 2.6) to have required treatment following injury in violence (see table 2 for a breakdown of ED attendance and rate of injury by year and gender).
Disaggregation by age group revealed higher rates of violence-related injury for males across all five age groups studied: those aged 18–30 years experienced the highest injury rates per 1000 population (males 18.4, 95% CI 15.6 to 20.3; females 6.3, 95% CI 5.4 to 6.9), followed by those aged 11–17 years (males 8.9, 95% CI 7.8 to 10.4; females 3.7, 95% CI 3.3 to 4.3), 31–50 years (males 6.8, 95% CI 5.9 to 7.6; females 2.9, 95% CI 2.5 to 3.2), 51 years and over (males 1.38, 95% CI 1.21 to 1.50; females 0.56, 95% CI 0.49 to 0.63) and those under the age of 10 years (males 0.41, 95% CI 0.31 to 0.50; females 0.22, 95% CI 0.16 to 0.26); ranking similar to those in the previous 10 years.13–15
Table 3 shows injury rates by gender, age group and region. Analysis by region revealed substantial variation between northern and southern regions, with higher injury rates identified in the North West (6.8/1000 residents, 95% CI 5.8 to 7.8) and North East (6.5/1000 residents, 95% CI 4.8 to 7.1). Lower rates of injury were identified in the Eastern (2.5/1000 residents, 95% CI 2.0 to 3.0) and South East (3.1/1000 residents, 95% CI 2.5 to 3.9) regions. Breakdown of injury rates by gender, age group and region revealed that although Yorkshire and Humberside had experienced a higher overall violence rate, rates of violent injury among children in this region were lower than those in Wales and London. Similarly, although London had only the fifth highest overall injury rate in England and Wales, it had the highest rate of violent injury among those aged 51 years and over and the third highest rate among adolescents.
National and regional trends and seasonality
Table 4 shows average annual trends in violence-related attendances at EDs by gender and age groups in England and Wales over the 5-year period. Violence affecting males and females decreased substantially, with an estimated average annual decrease of 13.8% (95% CI −14.8 to −12.9, p<0.05) between 1 January 2010 and 31 December 2014. Violence affecting all age groups studied decreased over the study period, with the highest average decreases among 11–17 (−22%, 95% CI −23.3 to −20.6, p<0.05) and 0–10 year olds (−19.5%, 95% CI −21.3 to −17.7, p<0.05) and the lowest average decrease in those aged 51 years and over (−8.3%, 95% CI −9.9 to −6.7, p<0.05).
Average annual trends in violence-related ED attendances by region are shown in table 5. Decreases in violence-related ED attendances were identified in seven English regions and in Wales over the 5-year period, with the most substantial decline in violence experienced in the West Midlands region. Regional trends by gender and age group are not presented here in detail but examination of these trends showed that the regions which experienced the greatest decreases in violence (ie, West Midlands, North East, Wales, South East and North West), did so across all gender and age group combinations.
Overall, violence-related ED attendances were highest in May (4.7/1000 population, 95% CI 4.0 to 5.2) and July (4.7/1000 residents, 95% CI 4.1 to 5.1) and lowest in February (4.0/1000 population, 95% CI 3.3 to 4.3). There was little difference between seasonal variation of male and female ED attendances over the 5-year period. Similar to the previous 5 years, the largest rate ratio between male and female ED attendances was identified in December (3.0).
This national study, based on a sample of 151 EDs, stratified by 10 regions in England and Wales, revealed significant decreases in violence affecting both males and females in the 5 years, 1 January 2010–31 December 2014. All age groups studied showed substantial decreases over this period. According to the NVSN data, violence-related injury in England and Wales which led to ED treatment declined by an average of 13.8% per year between 2010 and 2014 and, apart from a few single year increases, continues the steady national decline since 2001. The longer term reduction in violence according to the NVSN is consistent with the CSEW; numbers of CSEW incidents of violence declined by 21% between 2009/2010 (April–March) and 2013/2014.4 In contrast, the number of offences of violence against the person recorded by the police decreased by 14% between 2009/2010 and 2012/2013. However, violence according to the CSEW and NVSN continued to decrease between 2012/2013 and 2013/2014, whereas police recorded violence rose by 6%.4 Police records are greatly influenced by targeted policing and recording practices. For example, the installation of public space closed-circuit television (CCTV) cameras for crime reduction purposes led to an 11% increase in police detection of violence, but a 3% decrease in violence-related ED attendances.18 For these reasons, police records should not be used to measure trends in violence.19
In this study, likelihood of sustaining injury in violence varied by gender, age group and region: rates of violence-related injury were consistently higher for males, young adults (18–30-year olds) and in northern regions. Demographic comparisons with other violence measures are difficult; police recorded incidents of violence fail to disaggregate by age or gender, while CSEW violence data are only available at national level. Conclusions from the CSEW are however consistent with the findings of this study. Reasons for demographic variations in violence are likely to be multifactorial and complex but may include violence as a means to establish a strong masculine identity, higher levels of alcohol consumption among young adults compared with other age groups and north–south inequalities in health and prosperity. For example, it is known that intoxication increases the risk of injury by rendering people less physically capable, less likely to make sensible decisions in high-risk environments and more likely to walk home alone.20 ,21
Reasons for the national decline in violence-related injury and for regional variation over the study period are not clear. However, there is increasing evidence to suggest that public health interventions may be contributory. Information sharing between health services, police and local government implemented nationally following the 1998 Crime and Disorder Act and particularly after the tackling knife crime strategy was implemented in 2008, have been shown to substantially reduce violence-related injury.22–24 Both the extent and quality of information sharing for the purposes of violence prevention, between EDs and police, local authorities and other agencies differ between regions. A 2011 audit of information sharing partnerships in England, for example, revealed that, of the Community Safety Partnerships (CSPs) sampled, only 37% were using the data provided by EDs for violence prevention purposes; ranging from none of the CSPs in the East Midlands and North East regions to 56% of CSPs in the South East, respectively.25 According to a further audit, the proportion of CSPs using ED data had risen to 53% in 2014.26
Alcohol consumption trends may have also impacted on the national decline in violence-related injury. For example, hospital admissions for alcohol-related violence in England fell by 27% between 2005/2006 and 2013/2014 at a time when alcohol consumption (per capita) declined in the UK by 17% and heavy episodic drinking (more than 8 units per session for males and 6 units per session for females) among 16–24 year olds decreased by 11%.27–29
Seasonal variations in violence were clear. Overall, rates of violent injury were higher in May and July and lower in February. Similar to findings from previous studies, evidence indicates that violence is more common in the late spring and summer compared to other seasons.15 This may relate to longer daylight hours which result in larger numbers of people in urban centres for longer periods.30 Prevention efforts should focus on reducing violence during these months when risk of sustaining injury in violence is greater. On this basis, prevention and alcohol campaigns in the Christmas/New Year period seem less important now than in past decades but should be implemented in the period May–July as a priority.
Bias in ED data
ED attendances are not primarily recorded for the purposes of measuring trends in violence and, reflecting this, using ED data for this purpose has limitations. ED treatment depends on the presence of injury serious enough to require medical treatment and, therefore, ED data do not account for violence where no injury or very minor injuries are sustained. ED data are potentially liable to respondent bias; for example, victims may be unwilling to report that injury was sustained in violence. Alcohol intoxication can impair long and short-term memory and intoxicated patients may not be able to recall the circumstance of their injury.31 Variations in ED accessibility have also been shown to affect likelihood of attendance; particularly in more rural settings far from an ED.32
What is already known on this subject
National crime surveys and police records are contradictory in relation to trends in violence; neither measure allows conclusions to be drawn on demographic trends or violence seasonality.
Emergency department data provide an objective, harm-based measure of violence that can be used to triangulate measurement and bring clarity to national and local violence levels and trends.
What this study adds
National trends in violence according to emergency department records are similar to trends from national crime survey data but not trends derived from police records.
Violence-related harm in England and Wales decreased over the 5 years, with significant decreases in all but the Eastern and South West regions.
Higher violent injury rates were found in May and July and in the North West, North East and Yorkshire and Humberside regions.
Violence prevention efforts should be focused in regions with higher injury rates, in regions where violence has not decreased and in the late spring and summer.
The authors would like to thank the clinical leads of all the 151 EDs that took part in this study (see figure 1).
Correction notice This article has been corrected since it was first published Online First. The correspondence details have been changed.
Contributors VS, JS and KM designed the study. VS and JW contacted the clinical leads and collected the data. SM and NP cleaned and aggregated the data. PM and NP conducted statistical analysis. VS and NP prepared the first draft of the manuscript. All authors contributed to the main content of the manuscript and provided critical comments on the final draft. All authors have read and approved the manuscript prior to submission.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Aggregate anonymised data on violence-related emergency department attendances used in this study were obtained using Freedom of Information requests. Data were accessed under the agreement that aggregate data would not be shared with external parties. For this reason, unpublished data is not available on request.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.