Observatories for violence and injury

The model relies on collaboration between data stakeholders, government and non-government actors within the violence prevention and safety cluster.

The observatory seeks to collate and integrate all violence and injury-related data and not limit itself to a clinical perspective, but rather incorporates a public health approach which considers a broader contextual understanding of the data with regard to forensics (unnatural deaths), violence-related crimes, victim of crime surveys (unreported crimes), emergency medical services (EMS) violent incident data, structural correlates of violence (census data) and, finally, non-fatal trauma cases through hospital clinical database.

As described in detail previously,6 7 key functions of an observatory include

1. Observatory function 1: collection, integration and storage of secondary data and information.

2. Observatory function 2: data analysis.

3. Observatory function 3: reporting on and disseminating information and knowledge.

Historical background of the VIO model

Between 1993 and 1996, the mayoral administration of Cali, the third most populous city in Colombia, instituted a programme of development, security and peace, referred to as DESEPAZ.8 This programme applied a public health perspective to issues of violence prevention and intervention, influenced by the mayor’s background in epidemiology, and would establish the framework for the first ever observatory dedicated to the theme of violence and injury.8 The information was subsequently validated, supplemented and used in weekly meetings of the city’s Security Council, whose primary focus was citizen security issues; however, the council also sought to improve the coordination and efficiency in the use of resources.8

Following a thorough review of the data, further statistical analysis led to subsequent policy planning and coordinated intervention efforts by civil authorities.8 Concurrently, structural interventions to improve police functioning through the provision of pay increases, educational opportunities and housing construction incentives were implemented.9 These initiatives would provide the initial framework for later developments with the VIO model.

In the area of international violence prevention approaches, there has been a growing interest on the part of governments, municipalities, research centres, civil society organisations and international organisations in creating observatories or analytical tools for security-related problems, including, but not limited to, school violence, domestic violence, drug use and social and gender violence.10 Table 1 7 provides a sample of VIOs found globally.

Type of studies

We included non-randomised controlled trials (RCTs), quasi-experimental designs, prospective and retrospective cohort studies, controlled before-and-after studies and cross-sectional studies. Studies performed in general or specific populations and in hospitals or clinics were included. Additionally, we sought to include studies performed in any country and published in any language.

Types of participants

Participants for this study included adults ≥18 years of age who are located within the catchment areas of the observatory study sites.

Types of interventions/exposures

For purposes of the systematic review, we have used the term ‘observatory’ to denote a surveillance system that collects data from multiple sources, for example, crime data from policing sources and injury data from clinical and forensic sources, whereas injury surveillance systems (ISSs) almost exclusively focus on the use of injury data alone from clinical sources. We included observatories/ISSs that address violence prevention and whether these reduce violence in adult populations. All surveillance systems that focus specifically on the collection of violence and injury data were included in this review.

Types of comparisons

Controlled populations were extracted as presented in the respective articles. Control data were drawn from preintervention figures as specified by the authors. Where no data were supplied or documented, we extrapolated preintervention information from cities with a similar population size and make-up within the surrounding regions.

Types of outcome measures

Violence is defined as the intentional threat or use of physical force against oneself, another person or a group or community that results in injury, death, psychological harm, maldevelopment or deprivation.12 The outcome measures are based on the Organisation of American States (OAS) regional system of standardised indicators in peaceful coexistence and citizen security,13 as they represent the largest member organisation of crime and violence observatories worldwide, and included measures obtained by administrative record or surveys.

Primary outcomes

Primary outcomes included murder/homicide, suicide, transit death, unintentional injury death, sexual violence and intrafamily/family/domestic violence.13

Secondary outcomes

Secondary outcomes included aggravated assault, crime victimisation and the perception of insecurity, fear or risk.13

Patient and public involvement

Patients and/or the public were not involved in this research.

Database

We searched the following electronic databases: PubMed, Sociological Abstracts and International Bibliography of the Social Sciences and Education Resources Information Centre via Proquest, PsycINFO and Cumulative Index to Nursing and Allied Health Literature and Humanities International via EBSCOhost, SCOPUS, Cochrane Collaboration, Campbell Collaboration, Social Care Online, National Criminal Justice Reference Service, Web of Knowledge and Regional databases of the WHO. The searched Spanish databases included Spanish National Research Council, Epistemonikos, Evidence-informed policy network, Universidad de La Rioja, Red de Revistas Científicas de América Latina, el Caribe, España y Portugal, Scientific Electronic Library Online and the Virtual health library.

Furthermore, the following websites were searched for relevant literature: websites of the WHO Violence Prevention Alliance (http://www.who.int/violenceprevention/en/), Blueprints for Violence Prevention (http://www.colorado.edu/cspv/blueprints), the Community Guide (http://www.thecommunityguide.org/violence/index.html), Centres for Disease Control and Prevention (http://www.cdc.gov/ViolencePrevention/index.html), The World Bank (http://www.worldbank.org), the Juarez violence and injury observatory (http://observatoriodejuarez.org/dnn/ENGLISH.aspx) and the Medical Research Council (MRC) burden of disease research unit (http://www.mrc.ac.za/bod/bod.html).

In addition, the following conference proceedings were searched for relevant abstracts: International Conference on Crime Observatories, United Nations Congress on Crime Prevention and Criminal Justice, Global Violence Reduction conference, Annual Meeting of Violence Prevention Alliance and the International Society for Violence and Injury Prevention international conference. We used both text words and medical subject headings terms. The terms were used in varying combinations. Reviewers also searched reference lists of the relevant studies identified.

Selection of studies for inclusion

Review authors used a screening guide developed by AJ to ensure that inclusion criteria are consistently applied. Two review authors (AJ and DB), working independently, screened the titles and abstracts of all studies identified through the English literature searches for eligibility. MC and FF completed the same process for the Spanish language studies. Full texts of potentially eligible studies were obtained by AJ. The two authors (AJ and DB) independently assessed the full text of each article for eligibility and compared their results. Again, this was repeated by MC and FF for the Spanish language studies. Discrepancies were resolved through discussion and consensus, consulting a third author (MEE) to resolve any persistent disagreements. All reviewers documented the reasons for all studies excluded from the systematic review.

Data extraction and management

Two authors independently extracted descriptive and outcome data for each included article using a standardised data collection form, resolving any discrepancies by discussion and consensus, failing which a third author (MEE) would arbitrate.

Review Manager V.5.1 statistical software (http://ims.cochrane.org/RevMan) was used by AJ, while DB cross-checked the data entered to ensure that there were no data entry errors. References were managed using Mendeley V.1.19.3.

Incidence rate ratios (IRRs) were calculated from the extracted data.

Assessment of risk of bias in included studies

Two reviewers assessed all included studies using the Effective Public Health Practice Project (EPHPP) questionnaire, which is a quantitative study assessment tool to identify methodological issues.14 The criteria used to assess the risk of bias in RCTs were random sequence generation; allocation concealment; blinding of participants and study personnel; blinding of outcome assessors; incomplete outcome data; selective outcome reporting; other sources of bias and overall risk of bias, in accordance with the methods used by the Cochrane Collaboration and the EPHPP tool. The criteria used for risk of bias assessment for non-randomised studies include selection bias (dealing with confounding, adjustment and comparability of groups); performance bias (in terms of the fidelity of the interventions); detection bias (regarding unbiased and correct assessment of outcomes, including blinding of assessors); attrition bias (with regard to completeness of sample, follow-up and data); and reporting bias (with regard to publication biases and selective reporting of results).14 Studies were scored as having low, high or unclear risk of bias. Any disagreements between the two authors in the assessment of risk of bias were resolved in discussion and consensus and the consultation of a third author where necessary.

Data synthesis including assessment of heterogeneity

Data analysis was managed using STATA statistical software.15 Meta-analyses using the metan routine (metan logirr selogirr) were conducted by combining IRRs defined as the incident rate of violence outcomes (homicide or assault) before and after the introduction of a VIO or ISS. The random effects model was used. Heterogeneity was assessed by examining types of participants, interventions and outcomes in each study. Statistical heterogeneity was assessed using the χ² test and quantified with the I² statistic.16 Where heterogeneity was apparent, findings are discussed as a narrative summary.

Subgroup analyses

Subgroup analyses examining the VIO model (comprising aggregated data from regional clinical centres and mortuaries for homicide data) versus the ISS model (public hospital data) were performed.

Assessment of quality of evidence

We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to assess the quality of evidence for the contribution of the observatory towards violence prevention.17 The GRADE approach specifies four levels of quality ranging from high to very low, with the highest quality rating denoting a confidence that the true effect lies close to the estimate of the effect. Quality was rated according to an a priori identification of potential participant-centred outcomes, including benefits and harms.17 Two authors independently assigned the grade scores and compared results as per the process for the recording of previous aspects of the study. Discrepancies were resolved by consensus discussion between the two primary reviewers (AJ and DB), with arbitration by a third reviewer (MEE) as necessary.

Sensitivity analyses

We performed sensitivity analyses to determine the effect, if any, of type, and quality assessment of publication on outcome.

Description of included studies

The study selection process identified 3105 potentially relevant unique citations from all literature searches (figures 1 and 2 detail the results from English and Spanish databases, respectively). Of the 21 studies deemed potentially eligible for inclusion, 9 English language studies and 7 Spanish technical reports met the inclusion criteria. Three studies were conducted in the UK, with the remaining studies taking place in Colombia, Brazil, Mexico and Uruguay. Data from technical reports were used in the five studies arising from South America. All studies were published after 1990. No studies were RCTs. Control data were reported in all the empirical studies in the form of preintervention figures. None of the technical reports had control data, and thus estimates were extrapolated from World Bank open data.18

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram for studies from English databases.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram for studies from Spanish (n=3)/Portuguese (n=1) databases.

Details of studies included in the review are presented in table 2. Reasons for the 10 studies excluded from the systematic review are detailed in table 3.

Meta-analysis across all studies was not possible, given the heterogeneity in both the model and the nature of outcome reported. Subgroup analyses were thus conducted according to the two types of models implemented (table 4): three empirical studies used assault count as an outcome, while four technical reports used homicide rate. One technical report,19 initially deemed eligible for meta-analysis, was considered an outlier after closer inspection, and subsequently, excluded from the meta-analysis; instead, it is discussed in a narrative review below. The quality assessment of the included studies is described in table 5.

Overall effect of violence interventions

The pooled crude IRR for the seven violence intervention studies was 0.18 (95% CI 0.05, 0.71) (figure 3). Heterogeneity was observed across studies (I² statistic 100%).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Violence outcome count after VIO implementation. VIO, violence and injury observatory; ISS, injury surveillance system.

Subgroup analysis by model of surveillance system indicates an effectiveness for the introduction of a VIO in reducing homicide (IRR 0.06; 95% CI 0.02, 0.19) (four studies, n=44 043 495). In the three studies employing an ISS model, the pooled IRR for the association between the intervention and assault count was 0.80 (95% CI 0.71, 0.91). Sensitivity analyses for language of publication, region and type of report rendered identical results.

Narrative review

Among the six studies included in this review but not included in the meta-analysis, there was no control group and thus a statistical effect could not be determined. An additional limitation to these studies, including Arnetz et al,20 Gutierrez-Martinez  et al,8  Garrib,²¹ Mberu,²² Ward   et al ²³ and Zhang et al ²⁴ was the lack of preintervention data. Furthermore, Mberu et al 22 and Garrib et al 21 used data from demographic surveillance systems which were data collection tools whose primary focus was not violence-related injuries or outcomes. In the Arnetz 2011 study, the ISS implemented was limited to a large metropolitan multisite hospital system (six hospitals) whose sole focus was the recording of violence incidents that resulted in injuries to patients or hospital employees, thus limiting the generalisability of the results.

With regard to the study setting, each of the six studies had divergent settings. Arnetz et al 20 had a focus of violence among the hospital staff population within several regional hospital, while Ward et al ²³ was set in a single public hospital. Zhang et al’s24 focus was national, while Gutierrez-Martinez et al 8 focused on six municipalities within Colombia. Mberu et al ²² looked at an urban slum population on the outskirts of Nairobi, Kenya, while Garrib et al ²¹ was set in a northern KwaZulu Natal rural population in South Africa.

In terms of one of the primary outcomes, homicide count, only Garrib et al 21 and Gutierrez-Martinez et al ⁸ reported this outcome of these six included studies which did not undergo meta-analysis. Furthermore, when considering the secondary outcome, assault count, only Mberu et al 22 and Zhang et al 24 reported this outcome.

Finally, with regard to study design, the lack of a control group was a consistent feature for all six studies.

Comparison with previous systematic reviews and observational studies

To date, this is the first systematic review to assess the evidence of effect for VIO/ISS reduction of violence among adult populations. This review confirms the findings of previous observational studies, which have shown reductions in violence outcomes10 11 after the implementation of a VIO.

Strengths and weaknesses

One of the strengths of this review is that it incorporated a range of search approaches among a number of well-known databases, without language restriction, to ensure a comprehensive strategy in evaluating the evidence. In cases where English abstracts of publications in non-English journals deemed relevant, full-length foreign language articles were translated and assessed for inclusion in this review. The lack of evaluation of randomised controlled studies is a weakness generally noted in this type of research, which involve a prohibitive cost to run. Furthermore, the lack of controls, preintervention data, blinding and the reporting of withdrawals and dropouts were weaknesses identified within the included studies. A publication bias was noted in the South American literature which more often published data in regional technical reports, favouring this type of publication over peer-reviewed journal publication.

Interpretation and mechanisms

The formation of an observatory allows for the collation and integration of violence-related data. Furthermore, it brings together relevant stakeholders from government; emergency services including medical, police and forensic; and non-government actors in the safety sector to promote a comprehensive understanding of the burden of violence-related injuries through data sharing. The observatory practices include monitoring and evaluation of violence-related injuries and crimes, and intervention development with stakeholders to promote violence reduction. Interventions may be applied at multiple levels including places, people or behaviour-based approaches. Examples of a ‘place-based approach’ would include interventions focused on geographical locations such as hot spot policing or disorder (broken windows) policing. Examples of a ‘people-based approach’ includes ones focused on individuals and groups and include interventions such as focused deterrence, cognitive–behavioural therapy or vocational training. Finally, examples of a ‘behaviour-based approach’ include the targeting of known risk factors for violence such as firearms, alcohol, drugs and gangs. These may be addressed through policy changes or interventions such as drug courts and treatment.

An alternate consideration of the mechanism it is that observatories are the ‘outcome’ of an intersectoral initiative by various stakeholders to act collectively against violence and it is this collective action that brings about the reduction with the observatory being a mediator on that causal pathway.

Clinical relevance

VIOs and ISSs remain a viable model for the surveillance of violence-related injuries and outcomes. This is consistent with the public health surveillance principles which have been applied successfully to other communicable and non-communicable diseases worldwide. Physicians should be encouraged to share their violence-related data with violence prevention stakeholders, as this provides a more complete picture of the local burden of violence-related injury. This is demonstrated with the advent of targeted policing and the Cardiff model which promotes data sharing between hospital emergency departments and local policing services, and has demonstrated both reductions in hospital admissions related to violence and reduced the number of violence-related crimes reported to the police.21 29