Preventing violence in New South Wales: data sources and their adequacy

Public health efforts to prevent violence rely on information about communities most at risk of violence and the success (or otherwise) of current interventions. The World Health Organization World Report on Violence and Health recognised that information and data concerning violence are generally incomplete and inadequate for the purpose of public health action, and called for countries to enhance their ability to collect and present such data.¹ In this article we describe data sources that are relevant to violence in NSW and recommend strategies for improving data collection and dissemination.

As part of a larger project we interviewed 31 key stakeholders in government and non-government organisations in the health, community, criminology, police and other sectors, and asked them about what data sources they accessed for their violence prevention work. We distributed a short survey to custodians of 11 routine data collections managed within NSW Health that we identified as being relevant to violence; of the 11 custodians, nine responded. We also searched government websites and recent published literature to identify data sources.

Data sources identified were mapped using a conceptual framework that was developed by the Australian Bureau of Statistics (ABS) for mapping information relevant to sexual assault.² This framework characterises data sources according to whether they provide information before an incident (context and risk factors), at the time of an incident (incident) or after an incident has occurred (responses and outcomes). Some datasets provide information at more than one of these levels. This information is summarised in Table 1.

Table 1.  Data sources relevant to violence in NSW – conceptual framework adapted from the Australian Bureau of Statistics²

The context in which violence occurs includes individual factors, such as prior history of aggression; relationship factors, such as peer and family norms; community factors, such as social capital and residential mobility; and societal factors, such as historical and cultural circumstances. Risk factor studies, such as census data on employment and socioeconomic status, provide information on communities at known increased risk of violence, which can be used prospectively for planning interventions. Population level information, such as population density, drug and alcohol use and the status of women and children can be studied to determine relationships between these variables and violent incidents.

Example: The 2004 National Drug Strategy Household Survey³ found that one in three people aged 14 years and over consumed alcohol in a way that put them at increased risk of alcohol-related harm in the short-term on at least one occasion in the preceding 12 months. Risky drinking peaked in men and women aged 20–29 years.

Actual and perceived risks of violence are best documented through population surveys, including victimisation surveys, which are usually anonymous and can facilitate calculation of the prevalence of violence.

Example: The 2004 NSW Population Health Survey⁴ asked young people 16–25 years of age whether or not they had been victims of physical violence in the previous 12 months. Twelve and a half percent of young people had experienced a violent incident. The young men who had been physically assaulted were most likely to be assaulted in an outdoor place or a licenced premise, whereas young women who had experienced physical assault were more likely to be assaulted in the home.

Information at the incident level is primarily provided by routine datasets such as the NSW Inpatient Statistics Collection, NSW Recorded Crime Statistics⁵ and the ABS Mortality Collection. NSW Ambulance Service data could potentially provide information about violent incidents but are currently insufficiently sensitive for this purpose. The national Bettering the Evaluation and Care of Health (BEACH) general practice dataset includes information about violence coded according to International Classification of Primary Care (ICPC) codes, but the number of identified incidents of violence is very small.

Example: The NSW Inpatient Statistics Collection demonstrates that approximately 6000 people are admitted to hospitals in NSW every year for ‘interpersonal violence’ and males are three to four times more likely to be admitted than females.⁶

Data describing responses to violence by the health system include the National Minimum Data Sets collected through health service delivery points, such as drug and alcohol services,⁴ mental health services, trauma centres and child protection services and incident monitoring systems, such as the NSW Health Incident Information Management System, which documents reported violent incidents affecting clients or staff within NSW health services. In addition, specific mechanisms have been established in part to analyse system responses to serious incidents, such as the NSW Child Death Review Team⁷ and the Mental Health Sentinel Event Review Committee.⁸

Example: National data on patient-safety-related incidents from national incident management data demonstrated that from January 1998 to June 2002, 3621 incidents within the health system involved physical violence or a violent verbal exchange.⁹ The proportion of total incidents involving violence was higher in emergency departments and much higher in mental health units. These data were used to recommend strategies to train and protect staff in methods to deal with violent incidents.

The impacts and outcomes of violence are poorly described in current data collections. In particular, there is generally poor assessment of the psychological impacts of violence. Economic studies of the impacts of violence are important but have been conducted infrequently.^10,11

Example: The NSW Injury Risk Management Research Centre estimated the costs to NSW of various forms of injury during the period 1998–99.¹² The study estimated that self- harm was the second most costly form of injury to NSW ($588 million) and assaults were the fifth most costly mechanism of injury ($134 million). Falls are the most costly mechanism of injury at an estimated $644 million in 1998–99.

Table 2 provides an overview of the strengths and limitations of specific datasets relevant to violence managed by NSW Health, based on the comments of our informants, results of the survey conducted and a review by the UNSW team.

Table 2.  Routine data collections relevant to violence administered by NSW Health

Stakeholders interviewed were asked in general terms about the adequacy of the data sources they use. Some common themes emerged from these discussions:

Different data sources measure different variables, making comparison difficult. For example, sexual violence may be measured through data on presentations to sexual assault centres, presentations to emergency departments, police reports, court reports and population surveys measuring histories of sexual violence in childhood or adulthood. Sexual violence is defined differently in each of these data collections: for example, the legal definition of sexual assault used in criminal prosecution is much narrower than the definitions of sexual victimisation used in population surveys. Given that these collections are not linked, it is difficult to know how much overlap there is. One stakeholder from outside the health sector articulated the need for different sectors to share data more effectively:

‘The links between assault prevention, health outcomes and violence prevention are so clear, it is almost ridiculous that we are not sitting at a table on a monthly basis and comparing data…if anything can be done better it is the way we compare data’.

In concert with literature in the area, stakeholders interviewed perceived there to be significant under-reporting of intimate partner violence, sexual violence, elder abuse, school bullying and assaults on hospital staff. Some stakeholders were reluctant to identify violence if they perceived it would compromise provision of care or that there were inadequate referral mechanisms to follow up events identified: ‘We are not seeking to get individual level reporting of stuff around DV (domestic violence), child reporting…we wouldn’t know what to do with it’. Under-identification of Indigenous status and inaccurate reporting of violence in Indigenous communities were also reported.

The majority of victims of violence seen in the health sector are not admitted to hospital, but are seen in the emergency department or as outpatients. However, most data collections beyond the Inpatient Statistics Collection do not routinely record the intent of injuries, the external cause of injuries and the relationship between the victim and perpetrator. Although the Emergency Department Data Collection was identified as a key information source for improvement, this collection does not function primarily as a surveillance system and significant barriers to collecting good quality data in this setting were identified.

Several stakeholders outside the health sector reported that health data were difficult to access unless you knew where to look: ‘Health is such a complicated organisation from the outside’. One suggested that it would be ‘really good to have a contact in Health we can call up and say – right is it realistic for us to ask for this information – does it exist?’ Another thought that ‘A one-stop shop outlining the health data, who is the data custodian, how you get access to it…something like that would be beneficial’.

There are some communities ‘hidden’ in the statistics, such as gay and lesbian victims of both intimate partner and street violence, as most datasets do not record information on sexual identity. One stakeholder felt very strongly that the health sector collects poor quality data on child health, including child abuse, as most information is collected via parents rather than from children themselves.

Public health surveillance should enable the definition of public health priorities; the characterisation of disease by time, place and person; the detection of unusual patterns of occurrence or epidemics; the evaluation of prevention and control programs; and the projection of future trends and health care needs.¹³ All of these are relevant to violence, which can be detected to some degree in routine datasets at the incident level. However, certain forms of violence are likely to be significantly under-reported,^14,15 and even large scale victimisation surveys have been unable to systematically document the experiences of some groups, such as adult survivors of sexual assault, people from non-English speaking backgrounds, sex workers and the homeless.²

All routine data collection systems have weaknesses; added to these are some specific limitations of data collected about violence. Particular issues are the lack of resources and personnel to focus on violence in particular, and the use of non-standardised terms and protocols which limit the comparability of results derived from separate data sources or data collection instruments.¹⁶

To facilitate a broad understanding of the complex patterns of violence within society, information from different sources and different sectors is required. The increased use of data linkage methods is likely to assist with making these data more useful. Table 3 summarises the range of information that is available to violence prevention practitioners in NSW. It is not sufficient to characterise violence using only the ICD codes for assault in the NSW Inpatient Statistics Collection. Basing policy on these measures of morbidity alone will significantly underestimate the burden of violence in NSW. As victims of violence have higher health care needs and consume more health resources than those who have not been victimised,^1,17 it is appropriate that health systems pay particular attention to identifying violence and target programs to where they are most needed.

Collecting information about personal experiences of violence highlights specific ethical and logistic issues that may not be relevant to other forms of information gathering. In particular, respondents should not be adversely affected or revictimised in any way by participating in data collection.² Routine datasets often collect data on violence as a by-product of their main purpose, and are therefore unlikely to compromise staff time or patient acceptability. However, as discussed previously, many victims of violence are unlikely to be identified unless more active methods of information gathering are instituted.

Table 3.  Key resources relevant to violence prevention in NSW

Current information systems to characterise violence in NSW are only partially effective. Stakeholders reported that data about violence in general were not representative, were of variable quality and were difficult to access. In addition, there appear to be few systematic mechanisms by which information about violence is regularly reviewed and fed back into policy processes. Key recommendations arising from this project are currently being formulated for consideration by NSW Health.

Public health research, advocacy and intervention depend upon accurate estimations of the burden of injury. Improving information systems is the first step in the public health contribution to reducing the burden of violence on society.