• 
• 

Three Part Question

Which [Emergency Department] [interventions] are most successful at [reducing community assault]?

Clinical Scenario

You are a newly appointed Emergency Medicine Consultant at a large inner-city hospital, and you notice a high incidence of assault victims attending your department, and you wonder which interventions might be helpful in reducing this.

Search Strategy

(“Accident and Emergency” OR ED OR “A&E” OR Emergency Service, Hospital [MeSH]) AND (Intervention OR Data OR Experiment OR Trial OR Test) AND (Assault OR Victim OR Violen*) NOT Domestic NOT Sexual NOT Partner NOT Dating NOT “Intimate Partner”

Search Outcome

PubMed (MEDLINE) (1970 to 2016 April 15): 4235.
11 papers were relevant to the study question (Table 1)
Exclusion criteria: Any papers discussing interventions aimed at children aged 18 years or less only and any papers not assessing the efficacy of an ED intervention alone.

Relevant Paper(s)

Author, date and country	Patient group	Study type (level of evidence)	Outcomes	Key results	Study Weaknesses
Becker et al 2004 USA	112 youth (12 to 20 years) who were hospitalized for violent injury. Intervention group = 43, Control group = 69	Retrospective, comparative double cohort study over 6 months. Intervention group had a Peer Intervention Specialist offering visits and supports post-hospitalisation for a violent injury. The control cohort was a group selected from 1998 based on providing similar demographics while not receiving the intervention.	Arrest rates and the criminal justice system	Youth in the intervention group were less likely to be arrested over a 6-month period (Odds ratio, OR 0.257, [0.054 to 1.223], p = 0.06..) compared to those in the control group. This was not significant for violence-related arrests. A significant reduction in criminal outcomes was noted for individuals in the intervention group as compared to the control group (OR 0.356, [0.094 to 1.345], p = 0.096). Controlling for the severity of injury suggested that this reduction was most significant for those who were less severely injured.	The randomization strategy was not acceptable: Attempting to match subjects post-randomisation will inherently be subject to bias. Some evidence, based on histories of arrest and nature of the violent attacks, suggest that the intervention group displays different characteristics to the control group. This has a secondary effect of reducing the validity of the odds ratios calculated between the two groups. Altering p-values to produce significant data reduces the validity of the results. It was necessary to compare the odds ratios pre and post-hospitalisation to calculate a statistically significant difference after the writers identified differing characteristics between the intervention and control groups. However, the pre-hospitalisation baseline suggested that the intervention group had increased violence-related arrest rates. The 6 month follow up was too short, and this was combined with a small sample size, which may account for the lack of significant results.
			Rates of rehospitalisation or death from assault	No significant differences were measured between the intervention and control groups.
Warburton and Shepherd 2006 UK	Assaults presenting to an inner-city ED over three nine-month periods over 1999 to 2001. The first period was prior to the initiation of the intervention, the second at the beginning of the intervention, and the third post-tweaking of the intervention.	Longitudinal controlled intervention	Assaults inside licensed premises	There was an overall increase in number of assaults in licensed premises over the study period for the intervention group (12%), but an overall decrease for the control group (7%). No significant change was recorded for rate of assaults per venue per period for either the control or intervention group.	Urban areas of Cardiff were used as control, where the intervention occurred in the city centre. The demographic and characteristics of violence are highly likely to be different in these two areas – including a stated reduction in licensed premises in the control area compared to an increase in the intervention area. A method based on the size of a licensed premise was used to predict the capacity in a number of premises, this may reduce the accuracy of the results. Combining two odds ratios regarding the intervention venues lead to the drawing of a conclusion based on the significance of the ED intervention. However, this is incorrect as the two odds ratios were provided from two separate interventions that relied on differing police presence but the same ED consultant intervention. Thus the combined odds ratio gives no information on the ED intervention.
			Street assaults	An increase in the number of street assaults was observed in the intervention area (34%), with a decrease in the control group (7%). A disproportionate number of assaults occurred in one street which had an increase in licensed premises – these variables were positively correlated.
			ED intervention	Following high profile ED consultant and police intervention at one club in particular, there was a reduction in the number of assaults (events ratio 0.46, [0.31 to 0.70]). The odds ratio, comparing the change to nine other control clubs, was 0.60 ([0.37 to 0.97]). ED and low level police intervention produced no significant decrease in assault in a second club. A pooled odds ratio of the two intervention clubs showed reduced odds compared to the control clubs (OR 0.61, [0.40 to 0.91], p = 0.014
Zun et al 2006 USA	188 individuals aged 10-24 years presenting to an ED as victims of interpersonal injury that was either life- or limb-threatening. Patients randomly assigned into control (92) or intervention (96) groups based on envelope selection. Intervention group were case managed for 6 months, with the control group simply receiving a brochure.	Randomised control trial.	Differences between control and intervention groups	The proportion of the intervention group self-reporting reinjury post-intervention (8.6%) was significantly different to that of the control group (20.3%), (p < 0.05). No significant difference was found for return visits to A&E or involvement in the criminal justice system.	Self-reporting leads to an inherent bias. No evidence of blinding. A large number of individuals were lost prior to the beginning of the study, and during, for a number of reasons. This could possibly describe a population bias. The decision to include in the study was based on a decision by the lead investigator, which suggests a possible opportunity for bias.
			Intervention impact on targeted behaviours	No significant impact of the intervention was found for violent delinquency, nonviolent delinquency, drug use or violent victimization.
Shibru et al 2007 USA	154 ED patients aged 12 to 20 presenting over 1998 to 2003 as victims of assault.	Retrospective, comparative double cohort study. Same intervention group from a previous study (Becker et al, 2004). One cohort had received a hospital based peer-intervention programme, the second cohort was a group of matched patients that could act as a control. This study is an 18-month follow up.	Risk of death, physical reinjury or rehospitalisation from intentional violence	between the two groups at follow up.	No evidence of blinding. Controls were selected from a database with the attempt of matching the demographic features of the intervention group. There is no evidence of how this was carried out, or if there was any attempt at randomization. High p-value for suggesting that subsequent violent criminal behaviours were reduced. Self-reporting of reinjury risk accompanies a significant confound due to a reporting bias that could be affected, for example, by a peer-intervention.
			Subsequent involvement in the criminal justice system	Significant risk reduction for persons in the peer-intervention programme (Relative risk 0.67, [0.45 to 0.99]). Subsequent violent criminal behaviours, measured through police data, was reduced by 7% in the intervention group (p=0.15). However, logistic regression analysis suggested that this effect was more significant for those under the age of 17.
Aboutanos et al 2011 USA	75 patients aged 10 to 24 presenting to a level 1 trauma centre as victims of assault	Randomised controlled trial. Control: Brief violence intervention (BVI) alone, Comparison: BVI with 6 months of Community Case Management Services (CCMS)	Reinjury	One patient from each group was reinjured at a 6 months, no patients were reinjured at 6 weeks.	No information on how the groups were randomized, with some data suggesting different demographics between the groups. No evidence of blinding. Only 20% of eligible patients were enrolled which lead to a small sample size and suggests a population or selection bias. Outcomes such as employment, incarceration, education and recidivism were not, and are unlikely to be, showing statistically significant differences at 6 months. Longer term follow-up is required.
			Risk factor evaluation	Lower alcohol consumption measured in BVI + CCMS when compared to BVI – CCMS at 6 weeks (RR 2.5, CI [1.2 to 5.4]). No significant difference at 6 months. BVI + CCMS and BVI – CCMS showed significant reductions for marijuana use at 6 weeks (48% for BVI, 28% BVI + CCMS), but not at 6 months.
			Hospital service utilization postdischarge	Proportion of patients who scheduled their own clinic appointments was significantly increased in both intervention groups, compared to a historical cohort from past computer data. No significant difference in ED utilization, or appropriate utilization between the two intervention groups.
			Community Service Utilisation	Significantly greater utilization in BVI + CCMS group compared to BVI alone.
Florence et al 2011 UK	Assault victims presenting to Emergency Departments in Cardiff and 14 ‘similar’ cities from 2000 until 2007, with the Cardiff model implemented in 2003	Non-randomised controlled trial. Intervention: Cardiff ED data. Control: 14 other cities	Assaults incident rate ratio (IRR)	Total assault IRR decreased after the implementation of the Cardiff model (IRR 0.79 [0.73 to 0.85]). Common assault (minor injury) increased after the implementation of the Cardiff model (IRR 1.38 [1.13 to 1.70]). Wounding assault (serious injury) decreased after the implementation of the Cardiff model (IRR 0.68 [0.61 to 0.75]). All in comparison to the mean of the other cities. All controlled for year and population changes.	Lack of randomization, but the potential confound was reduced by using average data from 14 similar cities, and considering and controlling for a full set of possible confounding variables. No evidence of blinding. Due to the lack of randomization, it is unclear that the Cardiff Model is causative of the decrease in violence observed.
			Rates of violence recorded by the police	The increase in violence recorded by the police in Cardiff after the implementation of the Cardiff model was significantly less than the increase in violence measured in the other cities.
			Hospital-related violence admissions	There was a significant reduction in hospital-related admissions after the Cardiff model was implemented (IRR 0.58 [0.49 to 0.69]). The IRR controlled for year and population changes.
Quigg et al 2012 UK	242796 ED violence attendances through 2004 to 2010	Observational Study in one Emergency Department in North West England looking at the effectiveness of data sharing.	Trend in intentional harm – chi-squared trend analysis	There was a significant decrease in intentional harm over the course of the study (35.6%, p<0.001). The decrease was greatest for 5 to 17 year olds, and those over the age of 65, but occurred in all age groups.	An observational study provides no information about a specific intervention. There was no information on, or attempts to account for, nationwide assault rates. Self-harm was included as part of intentional injury, and is therefore a potential confound.
			Admissions	Proportion of intentional injury admissions decreased (5.3%, p<0.001)
Boyle et al 2013 UK	Victims of violent assault presenting to an Emergency Department through 2000 until 2011, after initiating the Cardiff Model of Data Sharing in 2005	Observational study. Measured number of assault attendances, number of hospital admissions and number of violent crimes recorded by the police.	Number of assault attendances	Number of assault attendances for men rose over the first two years (2005 to 2007) and then dropped back to baseline (2007 to 2011). The increase over the first two years was attributed to an incorrect flow of data sharing, rather than the model failing to have the desired effect. Assault attendance for women did not seem to change over the study period. Hypothesis testing was not performed.	The nature of this observational study renders it difficult to draw valid conclusions about the effectiveness of the Cardiff model on affecting incidence of violent assaults. Lack of hypothesis testing provides an inherent difficulty when comparing to other interventions. Data do not account for the increasing population that was mentioned.
			Number of hospital admissions for assault	The absolute number of hospital admissions for assault saw no change over the study period. Hypothesis testing was not performed.
			Police data on recorded violent incidents	There was a significant decrease in the number of violent crimes with injury recorded by the police, but no significant decline was recorded in the absolute number of violent crimes.
Smith et al 2013 USA	141 patients presenting to a trauma centre for assault who were considered to be high-risk for recidivism.	Retrospective cohort study. High risk individuals enrolled in a wraparound care programme after being identified as having a high risk of violent injury recidivism	Recidivism rate	Currently 4.5% with the wraparound care project. Historical rates defined as 16%.	Inclusion criteria was based on risk reinjury – assessments by trained case managers, that invariably leads to a selection bias. Data published in 1995 was used as a comparison. Exposure level with the case manager was shown to be dependent on age, a potential confound when looking at the validity of the conclusions. 70% of the patients had suffered gunshot wounds – there is unlikely to be a similar demographic of injuries in the UK (for the purpose of this review). A retrospective cohort cannot assess the efficacy of the intervention itself. Regression to the mean should be considered for a group where only the ‘highest risk individuals’ are chosen. All individuals from assault should have been included to negate this phenomenon.
			Case manager exposure	Those who received moderate as opposed to little exposure with a case manager had an increased chance of success (Odds ratio 4.8, [1.6 to 14.9], p = 0.007), with an even bigger success rate if they had high exposure (OR 5.6, [1.6 to 20], p = 0.006)
Droste et al 2014	8 primary research papers	Systematic review of ED data sharing at reducing community level violence due to alcohol	Data sharing	All studies reported benefits. The heterogenous nature of the data sharing interventions does not allow summation of the identified evidence.	Only 4 of the 8 studies were relevant to assessing the effectiveness of the intervention, and they have been included here with defined limitations.
Carter et al 2016 USA	409 patients presenting to an Emergency Department as a victim of assault, living in a particular neighbourhood	Non-randomised controlled trial. Two neighbourhoods based on postcode. Addressee’s of one neighbourhood received the intervention, the other received the control. Intervention: 30-minute therapist-led session (Brief intervention, BI). Control: Resource brochure	Physical aggression	Violent aggression was significantly reduced in both the control and intervention group. Violent aggression was significantly less in the intervention group compared to the control group at two-month follow-up (Incident Rate Ratio 0.87 [0.76,0.99]). No significant change was found in the prevalence of violent aggression for the intervention group as compared to the control group.	Victims were significantly more likely to refuse participation from the intervention than the control group suggesting an effect of the selection bias. There is no statistical evidence that the reduction in incidence of violent assault for the intervention as compared to the control is significantly reduced from the baseline – thus there is no information on the effectiveness of the BI compared to the control. No attempt at randomization. No evidence of blinding. Self-reported measures can be easily influenced, for example, by a brief intervention. The 2% of the intervention group who reported receiving other violence intervention should have been excluded.
			Victimisation	There was a significant reduction in violent victimization within the BI group at the follow-up, but no significant reduction was found for the BI group as compared to the control group at follow-up.
			Self-efficacy for non-fighting	Self-efficacy for non-fighting increased significantly in the BI group at follow up. There was also a significant increase in self-efficacy for non-fighting in the BI group as compared to the control group (IRR 1.09 [1.02,1.15]).

Comment(s)

This search identified three potential areas for intervention: Data sharing, Brief intervention and Advocacy groups. Largely, there is only weak data for interventions tested. This can partly be explained by difficulties inputting and measuring complex interventions and ethical approval (not offering a violence intervention to a group of individuals is generally considered unethical). In terms of data sharing, as the review by Droste et al (2014) suggests, three papers reported a fall in violent assault rates. The Boyle et al (2013) paper did, however, initially report a rise in assaults that was attributed to an incorrect data flow. An important aspect of the Boyle et al (2013) paper was the successful implementation of the Cardiff model in a large inner-city Emergency department. The paper by Warburton and Shepherd (2006) was the only paper to suggest an increase in violent related assaults after implementation of the data sharing, but this is possibly due to the small length of time since intervention that was measured. There was some heterogeneity in the methods of data sharing, but the general principles involved sharing data on location and characteristics of the assault with Community Safety Partnerships with the main outcome of targeting police interventions to specific areas of high assault rates. The highest level of evidence comes from Florence et al (2011) who performed a non-randomised controlled trial using the Cardiff Model of data sharing. They found a decreased incident rate ratio for total assaults, which was due to a decreased incident rate ratio for serious assault despite an increase for common assault. This mapped into a decrease in hospital admissions for assault. Although none of the studies can definitively say that data sharing was the cause of the reduction in assaults, it seems illogical to suggest that a such a low-cost intervention that targeted police resources would not be sensible to implement at any Emergency Department. Warburton and Shepherd (2006), as a supplement to the data sharing intervention, sent ED consultants and the police to nightclubs with high assault rates to pressurize them into implementing their own interventions with the aim of reducing assaults. Their data suggest that police and ED consultant presence does have a positive impact on assault rates at these nightclubs. Whether this is a feasible measure for all EDs could be a topic for debate. One study (Carter et al, 2016) looked at the efficacy of a brief intervention – a 30-minute therapist-led session with the intention of teaching youth about behaviours that negatively influence goals and behavioural strategies to avoid violence. The results of this study were largely that the brief intervention seemed to increase self-efficacy for not fighting, but no significant reductions in victimization were found. The authors found that violent aggression was significantly less for the brief intervention compared to the control at follow-up, but failed to statistically compare this to the baseline for significance testing. Without strong evidence supporting brief interventions, and the monetary expense of a team of therapists, it does not seem sensible to promote the use of this intervention in UK Emergency Departments. There was a large heterogeneity between papers assessing the efficacy of ‘Advocacy groups’ which, in general, provided intervention subjects with a certain amount of ‘Community care’, aimed at reducing community risk factors that could lead to violent assault. Aboutanos et al (2011) compared the efficacy of 6 months of community care against a standard brief intervention. While the Community Service utilization was greater in the intervention group, there was no significant differences measured for reinjury rates. Two studies (Becker et al, 2004; Shibru et al, 2007) utilized ‘peer-intervention’, whereby the community case manager was a trained individual from a similar neighbourhood, or who had been injured by violence in the past. While no reduction in reinjury rates were found, the effect of the intervention seemed to be to reduce subsequent involvement in the justice system. This effect seemed greater in younger individuals (Shibru et al, 2007) or in those with less violent injuries (Becker et al, 2004). Zun et al (2006) suggested that community care could have an effect on recidivism rates when compared with historical rates, and that the community care was more effective with a greater case manager exposure. The highest level of evidence is offered by Zun et al (2006), whose randomized controlled trial showed a decrease in self-reported injury recidivism rates, but showed no decrease in return visits to A&E or involvement in the criminal justice system. With no evidence to suggest a decrease in assault-presentations at Emergency Departments, and the significant cost of community care, it seems illogical to suggest this as a method for decreasing assault-related violence for UK Emergency Departments.

Clinical Bottom Line

The evidence for ED-implemented assault reduction programs is very thin, and, as such, there is insufficient evidence supporting the implementation of any of these interventions. However, it seems logical to suggest that Data Sharing should be strongly considered for implementation, as it is a low-cost approach with some data suggesting positive correlations with reducing violent assaults. Most studies reviewed in this article utilized the ‘Cardiff Model’ of data sharing.

References

1. Becker, M.G., Hall, J.S., Ursic, C.M., Jain, S., Calhoun, D. Caught in the Crossfire: the effects of a peer-based intervention program for violently injured youth. Journal of Adolescent Health 2004; 177–183
2. Warburton, A.L., Shepherd Tackling alcohol related violence in city centres: effect of emergency medicine and police intervention. Emerg Med J 2006; 12–17
3. Zun, L.S., Downey, L., Rosen, J. The effectiveness of an ED-based violence prevention program. Am J Emerg Med 2006; 8–13
4. Shibru, D., Zahnd, E., Becker, M., Bekaert, N., Calhoun, D., Victorino, G.P. Benefits of a hospital-based peer intervention program for violently injured youth. J Am Coll Surg 2007; 684–689
5. Aboutanos, M.B., Jordan, A., Cohen, R., Foster, R.L., Goodman, K., Halfond, R.W., Poindexter, R., Charles, R., Smith, S.C., Wolfe, L.G., Hogue, B., Ivatury, R.R. Brief violence interventions with community case management services are effective for high-risk trauma patients. J Trauma 2011; 228–36; discussion 236–237.
6. Florence, C., Shepherd, J., Brennan, I., Simon, T. Effectiveness of anonymised information sharing and use in health service, police, and local government partnership for preventing violence related injury: experimental study and time series analysis. BMJ 2011
7. Quigg, Z., Hughes, K., Bellis, M.A. Data sharing for prevention: a case study in the development of a comprehensive emergency department injury surveillance system and its use in preventing violence and alcohol-related harms. Inj Prev 2012; 315-320
8. Boyle, A.A., Snelling, K., White, L., Ariel, B., Ashelford, L. External validation of the Cardiff model of information sharing to reduce community violence: natural experiment. Emerg Med J 2013; 1020–1023.
9. Smith, R., Dobbins, S., Evans, A., Balhotra, K., Dicker, R.A. Hospital-based violence intervention: risk reduction resources that are essential for success. J Trauma Acute Care Surg 2013; 976-980; discussion 980-982
10. Droste, N., Miller, P., Baker, T. Review article: Emergency department data sharing to reduce alcohol-related violence: a systematic review of the feasibility and effectiveness of community-level interventions. Emerg Med Australas 2014; 326–335
11. Carter, P.M., Walton, M.A., Zimmerman, M.A., Chermack, S.T., Roche, J.S., Cunningham, R.M. Efficacy of a Universal Brief Intervention for Violence Among Urban Emergency Department Youth. Acad Emerg Med. 2016