Article Text

Download PDFPDF

External validation of the Cardiff model of information sharing to reduce community violence: natural experiment
  1. Adrian A Boyle1,
  2. Katrina Snelling1,
  3. Laura White2,
  4. Barak Ariel3,
  5. Lawrence Ashelford1
  1. 1Emergency Department, Addenbrooke's Hospital Cambridge University Hospitals NHS Foundation Trust, Cambridge, Cambridgeshire, UK
  2. 2Clinical School, Cambridge University, Cambridge, UK
  3. 3Institute of Criminology, Cambridge University, Cambridge, UK
  1. Correspondence to Dr Adrian A Boyle, Emergency Department, Addenbrooke's Hospital Cambridge University Hospitals NHS Foundation Trust, Box 87, Cambridge, Cambridgeshire CB2 2QQ, UK; Adrian.boyle{at}


Background Community violence is a substantial problem for the NHS. Information sharing of emergency department data with community safety partnerships (CSP) has been associated with substantial reductions in assault attendances in emergency departments supported by academic institutions. We sought to validate these findings in a setting not supported by a public health or academic structure.

Methods We instituted anonymous data sharing with the police to reduce community violence, and increased involvement with the local CSP. We measured the effectiveness of this approach with routinely collected data at the emergency department and the police. We used police data from 2009, and emergency department data from 2000.

Results Initially, the number of assault patients requiring emergency department treatment rose after we initiated data sharing. After improving the data flows, the number of assault patients fell back to the predata-sharing level. There was no change in the number of hospital admissions during the study period. There were decreases in the numbers of violent crimes against the person, with and without injury, recorded by the police.

Conclusions We have successfully implemented data sharing in our institution without the support of an academic institution. This has been associated with reductions in violent crime, but it is not clear whether this association is causal.

  • violence
  • violence, interpersonal

Statistics from

Request Permissions

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.


Community violence is common throughout the UK, though there has been a steady, but slight decline in the numbers of patients receiving care for assault-related injuries over the last 10 years. Around 300 000 patients were treated for assault in emergency departments in England and Wales in 2011.1 The annual health costs of violence are estimated to be over £2 billion.2 The costs to the criminal justice system are also substantial.2

Reducing community violence is a political priority for the UK Government who have made a coalition commitment ‘we will make hospitals collect and share non-confidential information with the police so that they know where gun and knife crime is happening and can target stop and search in gun and knife crime hotspots’.

Work from Cardiff University and Liverpool John Moore's University has shown that information sharing of emergency department data can reduce community violence by around 30%. 3–6 The intervention is an active surveillance system, by which receptionists ask assault victims about the location of assault and record this in free text, the time of the assault and a description of the weapon was used. These data are collated in an anonymous form and shared monthly with crime analysts at the Community Safety Partnership (CSP). CSPs exist in most local authority areas. These are composed of representatives from local government, the police and the health service. The crime analysts produce a report containing the assaults known to the Police and those known to the emergency department. An emergency medicine Consultant attends the regular operational meetings of the CSP.

Research shows that emergency department data improves the quality of operational intelligence, as studies suggest that up to 75% of community assault patients needing hospital treatment are not reported to the police.7 The information within the reports allows the police to target their resources, define hotspots of crime, better deploy closed circuit television (CCTV) cameras and inform licencing decisions and appeals. Licencing decisions are made on a ‘traffic light system,’ which awards points for high rates of assaults and gives credit for actions likely to reduce assaults. The ‘traffic light’ system makes allowance for the size of the premises, so that larger premises are not unfairly penalised. This is generally known as the ‘Cardiff model for violence prevention’.

The considerable reductions in the numbers of hospital attendances for assault seen in Cardiff and Arrowe Park Hospital (supported by Liverpool John Moore's University) have been supported in part by an academic structure and as part of an ongoing research programme.6 Despite the evidence, uptake of this model of information sharing has been slow in other hospitals in the UK, despite being advocated by the College of Emergency Medicine.8 It is currently uncertain whether this model of information sharing can be successfully applied to other hospitals.

We aimed to validate this model of information sharing outside an academic setting. We aimed to reduce the number of assault victims attending the emergency department, reduce the number of admissions for assault, and improve police performance by adopting this model of information sharing. We aimed to describe the barriers in implementing this model.


We evaluated the model and its benefits within a natural experiment design in Cambridge, England. Our emergency department sees around 95 000 patients per annum. The hospital serves a mixed urban and rural catchment area and has two major universities. The city has an active night-time economy, based substantially on tourism and higher education.

In 2005, we initiated data sharing with the research team at the County Council. In most English cities outside London, local government administration is shared between a county and the city council. We instructed our receptionists to collect the three data items that the Cardiff Model require; a free text description of the location of the assault, the date and time of the assault and what weapon was used. We required a minor technology upgrade costing around £5000, to allow us to collect the data item on weapons. The anonymous data was submitted monthly and we monitored our performance by the number of assault patients who required emergency department treatment. In 2007, it became apparent that this model was not producing the desired effect and we re-evaluated the data flows. We found that the data we produced was not being sent on to the crime analysts at Cambridgeshire Constabulary, but was being sent to the research analysts at Cambridge County Council. We rectified this so that data was sent to the police crime analysts directly. A senior hospital manager, LA, invigorated the CSP and brokered increased acute hospital trust involvement with the CSP. The receptionists, who collect the data, receive feedback in the form of the crime reports, so that they could see the utility of the data they were collecting. This led to a virtuous cycle of data quality improvement. Initially, the receptionists collected usable data about location in about 20% of cases. This has improved and is now about 70%. The efforts of the receptionists were encouraged and recognised with a Community Safety Award from the police.

There were initial concerns about sharing potentially patient-identifiable data with the police when we initiated the project. This barrier was overcome with a previously issued statement from the Information Commissioner's Office. This stated ‘The Data Protection Act 1998 is not a barrier to the sharing of personal information. It should not be seen as preventing any trust sharing this data in a responsible manner.’ The Information Commissioner's Office also stated that attempts to repersonalise the data by the police would constitute irresponsible use of data by the police. The Information Commissioner's Office was supportive of the principle of data sharing for violence prevention. We obtained the support of our hospital Caldicott Guardian and data protection officer in developing an information-sharing agreement.

An emergency department consultant attends the bimonthly meetings of the Violent Crime Task Group, a subcommittee of the CSP. The Violent Crime Task Group is chaired by a senior police officer, and has representatives from the City Council, Licensing Authorities, CCTV operators, ambulance staff, various private companies providing door security at licenced premises and the street pastors.

Reassuringly, none of the licenced premises in our city centre lost their licences as a result of the Violent Crime Task Groups activities; however, a number of premises underwent reviews of their licencing conditions and had to submit an action plan.

We evaluated the effectiveness of our approach by interrogating the hospital computer system for the number of assaults and the number of assault-related admissions. We also examined the numbers of violent crimes, with and without injury, recorded by the police on the IQUANTA system. This is a police performance management system. This only goes back to 2009 with consistent case definitions.

We conducted hypothesis testing in STATA V.7, using the NPTREND command. There were no data to guide our sample size. We did not obtain ethical approval as this was a service evaluation.


The number of assault patients attending Addenbrooke's emergency department was considerably smaller than those attending Cardiff or Arrow Park Hospital, despite having similar numbers of emergency department attendances each year.3 ,6

There was an increase in the number of assault patients attending the emergency department from 2005 to 2007, after which there was a decline, returning back to a baseline (see figure 1). As expected, most assaults affected men. Men underwent the greatest decline in the number of assaults; the assault numbers were constant in women. The number of admissions was smaller and did not vary much over the study period (see figure 2). The non-linear shape of the curves indicated that hypothesis testing would be unhelpful and potentially misleading.

Figure 1

Number of assault attendances.

Figure 2

Number of hospital admissions for assault.

There was a decline since 2009 in the number of violent crimes recorded by the police, this trend was not statistically significant (z=−0.44; p=0.66) (see figure 3). At the same time, there was a statistically significant decline in the number of violent crimes with injury recorded by the police (z=−4.41, p<0.05) (see figure 3).

Figure 3

Number of recorded violent incidents against the person; Cambridge community safety partnership.


We have shown that it is possible to implement effective information sharing in our institution, with minimal extra funding or infrastructure development. We have shown a 20% reduction in the number of assault patients requiring emergency department care, and a 35% reduction in the violent crimes with injury reported to the police, but no decrease in admissions due to assault or violence. We have not shown a decrease in the number of admissions, as opposed to attendances, following assault.

It may appear counterintuitive that the number of hospital admissions did not decline, but may have risen slightly; while the number of police-recorded incidents of violence against the person with injury, seemed to decline. This apparent anomaly should be viewed in context. Hospital admission is a poor measure of assault severity, as admission practises vary depending on staff seniority, availability of CT scanning, observation wards and time-based performance targets.

This model of data sharing is being promoted by the Department of Health. Uptake within the NHS has been patchy. This is partly because there is no incentive system to encourage NHS hospitals to engage in injury prevention, but also because working across agencies can be complex. Injury prevention is a change of role for many clinicians who see their primary role as responding to injured patients.

We have not proved that the reduction in assault presentations to the emergency department and recording of violent injuries by the police is caused by emergency department data sharing. It is possible that there are alternative causes for the reductions in violent crime in our area. The economic crisis may have reduced the number of people going out and being assaulted. Nevertheless, the apparent association between the agreement and the reduction in violent crime is grounded, especially when taking into consideration that we have shown consistency with the two other UK sites that have successfully initiated data sharing and reported results; in all three sites, a substantial reductions in the number of assault victims attending their emergency departments was achieved.3 ,6 We have also shown a temporal relationship, in that assault numbers started to fall as soon as we started to send the data to the correct analysts, which strengthens our initial proposition that information sharing leads to a reduction in violent crime. However, it is scientifically impossible to identify from this study whether data sharing has caused reductions in the numbers of assault patients. There has been a national trend towards less assault victims attending emergency departments over the last 3 years, conversely, the study site serves an area with very rapid population growth. There were other interventions by the Violent Crime Task Group that might have led to reduced assault patients; dissecting out whether data sharing causes reduced community violence is scientifically challenging. Whether data sharing causes reduced community violence can only really be answered by randomised controlled trials.

Some practical utilisations are already visible following implementation. The emergency department data were instrumental in supporting the city council against a licencing appeal. A retailer applied to sell alcohol close to a homeless shelter, in an area designated as a ‘cumulative impact zone.’ These are areas that are known to have high rates of antisocial behaviour and alcohol-related problems. After an initial rejection, an appeal was lodged by the retailer. This appeal was, unusually, rejected, in part because of the emergency department data. We found that the police and licencing authorities were usually aware of problem premises and local violence hotspots, but that emergency department data provided added support in influencing problem premises and instituting control measures in violence hotspots. This finding, that emergency department data did not identify new geographical hotspots, differs from the evaluations from Cardiff and the northwest, where there are larger urban areas, and emergency department data assists identification of violence hotspots. We speculate that this is because the night-time economy is concentrated in a very small area in the city centre.

The emergency department data were also instrumental in identifying that foreign students were being assaulted on Monday evenings. It became apparent that there were irresponsible drinks promotions targeting these students on Monday evenings. The emergency department data allowed appropriate self-regulation by licenced premises. CCTV cameras were relocated on the basis of emergency department data.

There are some important limitations to our work. We do not have reliable denominator populations. In this context, we have found that around half our assault victims have postcodes out of the city, which makes defining the catchment area of the hospital difficult. Furthermore, Cambridge is a rapidly growing city, and our local population has expanded substantially over the study period, but the effect of this would be to minimise any benefits we have seen.

Another practical challenge has to do with our access to police data, which only goes back to 2009. We started sharing ambulance data about assaults with the police in 2011. Preliminary analyses show that 30% of ambulance callouts for assault do not result in transfer to hospital. The police do not invariably attend all ambulance callouts for assault, mainly as they are simply unaware of these emergency calls for service. We hope that this will further inform the work of the Violent Crime Task Group.


We have successfully implemented data sharing in our institution centre without academic support. This has been associated with reductions in both the numbers of patients requiring emergency department treatment for assault and violent incidents with injury recorded by the police, though we cannot state whether these are causal. This is consistent with previous work.



  • Contributors AB and LA initiated the project. AB wrote the paper. LA collected data. KS and BA commented on the manuscript.

  • Funding This work was unfunded.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.