Personal Security Alarms for the Prevention of Assaults Against Healthcare Staf Chloe Perkins1, Deirdre Beecher2, David Colas Aberg3, Phil Edwards1 and Nick Tilley3*

Perkins et al. Crime Sci (2017) 6:11 DOI 10.1186/s40163-017-0073-1

SYSTEMATIC REVIEW Open Access Personal security alarms for the prevention of assaults against healthcare staf Chloe Perkins1, Deirdre Beecher2, David Colas Aberg3, Phil Edwards1 and Nick Tilley3*

Abstract Background: Personal security alarms have been used to try to reduce violence against healthcare staf, some of whose members face relatively high risks of assault. This systematic review focused on the efect of alarms in reducing the incidence and/or severity of assaults. Methods: Electronic databases, including Cochrane Library, Ovid MEDLINE(R); CINAHL Plus (EBSCO); PubMed; PsycINFO (OvidSP) PsycEXTRA; Applied Social Sciences Index and Abstracts (ProQuest) (1987 to current); Criminal Justice Abstracts (EBSCOhost); Psychology and Behavioural Science Collection (EBSCOhost); Social Policy and Practice (OvidSP) Sociological Abstracts; ProQuest theses and dissertations, were searched. Study designs eligible for inclusion were randomised controlled trials, interrupted time series and controlled before-after studies that assessed the impact of personal security alarms on assaults. Searches were undertaken for studies of healthcare staf in all settings (i.e. including staf working in confned spaces such as hospitals and also feld personnel such as community health workers). Workplace violence between colleagues (lateral violence and bullying) and other uses of personal alarms (e.g. fall alarms for the elderly, domestic violence prevention) were excluded. Search results were screened by title, abstracts and keywords for possible inclusion. Full text reports for all potentially relevant studies were obtained and independently assessed for fnal inclusion. The primary outcome was physical assaults (recorded or self-reported). Secondary outcomes included increased confdence or self-efcacy in violence prevention (recorded or self-reported). Main results: No studies were found that met all inclusion criteria. Four reported associations of personal alarms (and other variables) with risks of assault in healthcare settings. These were described narratively. Conclusions: Healthcare workers in emergency departments, psychiatric units and geriatric facilities face much higher risks of assault than those in other healthcare settings. Alarm systems vary widely. Alarm systems form one of a range of measures, which may interact with one another, that are used to reduce the risks of assault. Given this complexity and diversity, prior to feld trials EMMIE orientated efcacy trials are recommended to try to establish whether alarms can be introduced and operated in ways that can contribute to reducing assaults in specifc high-risk settings. In relation to fndings relating to any given intervention, EMMIE refers to efects produced, mechanisms activated to produce the efects, moderators or contexts relevant to the activation of mechanisms, implementation issues that arise, and economic costs and benefts. Keywords: Violence against healthcare workers, Alarms, EMMIE, Systematic review, Crime prevention, What Works Centre for Crime Reduction

Background been assaulted and injured (Spector et al. 2014). In a 2010 Violence against healthcare staf is a major problem. An survey, between 5 and 8% of frontline National Health international review found that a third of nurses have Service (NHS) staf reported being physically assaulted by patients or other service users in the previous 12 months *Correspondence: [email protected] (Ipsos MORI 2010). Such assaults against NHS staf have 3 Department of Security and Crime Science, University College London, increased. According to NHS Protect fgures, a total of 35 Tavistock Square, London WC1H 9EZ, UK 70,555 physical assaults on NHS staf were reported in Full list of author information is available at the end of the article

© The Author(s) 2017. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. Perkins et al. Crime Sci (2017) 6:11 Page 2 of 19

the year 2015–2016 resulting in 1740 criminal sanctions; Te high risk of assaults for NHS staf is suggested to this represents an increase in total assaults of 4% from be due to factors including: inadequate security, staf 67,864 in the year 2014–2015 and an increase of 17% shortages, night-shift patterns, and the intensity of inter- from a total of 60,385 in the year 2004–2005 (NHS Pro- actions with patients (Oztunc 2006). Other contributory tect 2017a). Nurses are four times more likely to experi- factors reported include: patient mental health, patients ence assaults than any other NHS worker, with student under the infuence of alcohol or other drugs, waiting nurses and those in psychiatric and learning disability time and delays, problems understanding information, areas at highest risk (Wells and Bowers 2002). One quar- anxiety caused by practical issues (e.g. transport and ter of assaults are reported in acute wards and 69% in parking), and an increase in expectations of standards of those for mental health problems and learning disability. service which may not be met (Ipsos MORI 2010; Rew Globally, nurses working in some departments face much and Ferns 2005). higher levels of violence than those in others. Te pro- portions experiencing violence in geriatric wards (45.9%), Interventions emergency departments (49.5%), and psychiatric depart- Scoping literature searches revealed many studies that ments are substantially higher than those in a general identifed the problem of assaults against staf in the sample of nurses (23.4%) (Spector et al. 2014). Te source healthcare setting, and many studies that estimated the of these physical assaults against nurses have been found prevalence of such attacks and the exposure of healthcare to be predominantly patients, accounting for 64%, with a staf to violence. However, there is a paucity of primary further 30% being carried out by the family or friends of research into interventions to prevent violence. Interven- patients and the remaining 11% is lateral violence from tions to prevent violence against healthcare staf may be nurses, physicians or staf (Spector et al. 2014). broadly classifed as: Te Crime Survey of England and Wales identifes occupational groups at most risk of violence. Te 2014– •• Environmental (e.g. stafng numbers; CCTV; fxed 2015 Survey estimated that 1.3% of employed adults of emergency alarms; personal security alarms; secu- working age had experienced one or more incident of rity guards; police ofcers in A&E; making building violence at work over 12 months. It found workers in access more secure for staf; ambient environments, protective service occupations (e.g. police ofcers) to proper lighting and calming décor; facilities for chil- be at greatest risk with 9.2% experiencing one or more dren, e.g. play area); incident. Next came health and social care associate •• Practices and policies (e.g. legislation such as the professionals, such as most nurses, of whom 6.1% suf- Criminal Justice and Immigration Act 2008; ‘zero tol- fered one or more incident—over four times the overall erance’ approach policies); rate (Health and Safety Executive 2017b). Tose work- •• Staf skills (e.g. NHS confict resolution training; ver- ing in protective services and health have consistently bal techniques; prevention and recognition strate- been reported, over a number of years as being at high gies; staf attitudes, knowledge and skills) (Anderson risk. et al. 2010). In 2000, the Home Ofce estimated the total average cost of each common assault at £540, which includes the Key measures identifed by NHS Protect to tackle vio- physical and emotional impact, victim services, lost out- lence in healthcare settings include (Business plan 2014): put and the police court and prison costs (the average establishment of NHS Serious Incident Reporting System for more serious incidents of violence against the per- (SIRS) and Physical Assault Reporting System (PARS); son, excluding homicide, rising to £19,000) (Ofce 2017). confict resolution training (NHS Protect 2013); use of Impacts of violence on staf include pain, time away from the powers under the Criminal Justice and Immigra- work, depression and low self-esteem; impacts on the tion Act 2008 (CJIA) to tackle nuisance or disturbance NHS include loss of personnel for signifcant amounts of behaviour (sections 119–120 make an ofence of causing time and resignations (NHS 2003). Violence and abuse a nuisance or disturbance on NHS premises, and provide against staf present substantial costs to the NHS through power of removal); and guidance on the prevention and additional staf training, security, staf absenteeism, poor management of challenging behaviours in NHS settings. staf retention, and legal fees. Direct fnancial costs of A 2003 report from the National Audit Ofce identifed £69 million per annum have been attributed to physical a variety of security measures in place across NHS trusts; and non-physical violence and aggression in the NHS panic alarm systems (in 85% of trusts); closed-circuit (National Audit Ofce 2003). Te cost of physical assaults television (CCTV) (in 92% trusts); security staf (40%); during 2007–2008 was estimated at £60 million (NHS police presence (20%) and other measures (17%) includ- 2007). ing restricted access, security screens in reception areas Perkins et al. Crime Sci (2017) 6:11 Page 3 of 19

and mobile phones for lone workers (National Audit (linked to switch board and/or police) and portable Ofce 2003). Of the two most widely used security meas- personal devices (linked to central system with loca- ures, CCTV has been reviewed previously (Farrington tion information). Lone worker devices may also be et al. 2007; Welsh and Farrington 2009) and considered as included under this category. part of the What Works in Crime Reduction series (Col- lege of Policing 2015), while the use of personal alarms, Lone workers are a particular category of potentially at has received less attention. Personal alarms were there- risk workers, prevalent in healthcare and other occupa- fore, selected for further study as the focus of this review. tions, who work independently (often in a mobile capac- ity such as community healthcare providers but also The use of personal security alarms in healthcare settings independently or alone within buildings) without close Personal security alarms are used across NHS trusts. A supervision (Health and Safety Executive defnition). Health & Safety Commission report (Commission HaS NHS Protect specifcally defnes lone working as: 1997) outlined three types of alarm system: “any situation of location in which someone works without a colleague nearby; or when someone is •• Panic button systems are part of an internal alert sys- working out of sight or earshot of another colleague” tem, which often comprise hardwired buttons placed (p5 NHS Protect 2017b) in locations where there is a high risk of violence. Teir activation triggers an alarm on a monitoring Working alone is generally taken to increase a worker’s console. Tese may be useful in treatment and con- vulnerability, and the proportion of lone workers injured sulting rooms, where their location is known only to through physical assault is estimated to be 9% higher than members of staf (Commission HaS 1997). Clear pro- for non lone workers (NHS Protect 2017b) and is thought cedures on when and by whom they should be acti- to result in a greater severity of assault (NHS Protect vated should be set out and include a predetermined 2015). Lone worker devices or applications are covered response plan. Portable panic attack devices may by British Standard 8484:2016, a bench mark standard, also be linked to an internal system providing similar which Lone Worker Devices must reach for certifcation locational alerts on a monitoring console. and which forms the basis for a police response. Guid- •• Personal security alarms range from simple ‘shriek’ ance on lone workers and violence from the Health and devices, designed to shock or disorientate an attacker Safety Executive, further reports the most popular equip- to give victims time to get away, to a component in ment used by lone workers to be: a monitored system (as above). Simple audible alarm devices are not based on the expectation that they •• Mobile phones or other communication devices, to will produce assistance from third parties. Rather, call for help if needed and report their whereabouts they are primarily intended to create a distraction to and schedule to others, and allow the worker time to get away from a potentially •• Personal alarms, which can help staf feel more con- violent situation. NHS Protect also recommend that fdent about their safety (Health and Safety Executive the personal alarms are discarded after activation to 2017a). divert the assailant’s attention to silencing it (NHS Protect 2017b). Some experts advise personal audi- Alarms are seldom the sole means by which attempts ble alarms are more suitable for outdoor use due to are made to prevent violence against staf in healthcare the potential risk of escalating a situation indoors and settings. Te contexts in which alarms are used vary in their use now is more limited. It has been suggested terms of the other violence prevention methods in use, that alarm systems that rely on the use of whistles or clinical specialism, whether lone working is a factor and screams are inefective and dangerous in in-patient population served. Tis review will draw together all or psychiatric settings and coded, silent alarms trig- studies that examine the use of any personal alarm inter- gering a response (as with a panic button) are more ventions implemented to address the risks of violence desirable in this setting. Personal security alarm or and assault. noise devices have been highly recommended for For this review we searched for studies in electronic feld personnel (e.g. community health workers) databases and the grey literature primarily relevant to (Occupational Health & Safety Agency for Health- NHS or healthcare settings. However, to estimate the care 2005). efectiveness of all categories of personal security alarms •• Complex personal alarm systems include personal in healthcare settings, we broadened our searching to alarms linked to fxed detection systems e.g. by radio include evaluations in any other settings. We searched for or infra-red. Components may include panic buttons studies in any settings which may include both mobile and Perkins et al. Crime Sci (2017) 6:11 Page 4 of 19

feld personnel (e.g. paramedics and community health healthcare and other high-risk staf. Were sufcient num- workers) and those working in confned spaces (e.g. hos- bers of well-designed controlled evaluations identifed, pitals, mental health units, and GP surgeries) and did not we intended to include estimates of the efects on the restrict searches so that other high risk occupations could defned outcomes of interventions using personal secu- be included if relevant studies were found (such as trans- rity alarms. We intended to describe: participant charac- port workers and lone workers). In practice, we found teristics, setting, recruitment methods, theoretical basis none in non-health occupational settings, which is why used in designing interventions, intervention implemen- this review is confned to alarms in healthcare contexts. tation, delivery, and outcomes. Information speaking to the EMMIE framework was EMMIE framework extracted from the evaluation studies (Johnson et al. Tis systematic review was conducted in support of 2015). For each study we intended to describe the setting, the What Works Centre for Crime Reduction, hosted by theoretical basis for the intervention, characteristics, and the UK College of Policing. One aim of the UK College outcomes. We planned to summarise costs of the pro- of Policing is to promote and facilitate evidence-based grammes where economic data were available. policing, defned as: Methods “a method of making decisions about ‘what works’ Criteria for considering studies in policing: which practices and strategies accom- Broad search strategies were used in order to include plish police missions most cost-efectively” (Sherman evaluations of personal alarms in all settings that may 2013) (p. 377). be relevant to healthcare. Studies had to report on pro- Systematic reviews of the research literature lie at the grammes where there had been controlled evaluation of heart of the ‘what works’ movement. Te results of this personal alarms. review will be incorporated in an online toolkit devised by researchers at the UCL Jill Dando Institute of Security Types of studies and Crime Science. To help structure the toolkit, the con- Experimental study designs were to provide evidence sortium produced the ‘EMMIE’ framework for assessing of efects, and included controlled-before-after studies, fve dimensions of evidence (Johnson et al. 2015). Te controlled interrupted time series, controlled trials and EMMIE acronym refers to: randomised controlled trials that assessed the impact of personal security alarms on measured outcomes. •• Efect size (how efective is the intervention?). •• Mechanism (how does the intervention work?). Types of participants/populations •• Moderators (in which contexts does the intervention Inclusion criteria work?). •• Primary populations of interest were healthcare staf •• Implementation (what is needed to implement the in all settings (i.e. including staf working in confned intervention?). spaces such as hospitals as well as feld personnel •• Economics (how much might the intervention cost?). such as community health workers); •• Secondary populations of interest included work Tese dimensions encapsulate the types of evidence place violence in other settings with high risk of vio- studies might provide which could inform improved deci- lence (such as some council departments, local hous- sion-making (Sidebottom et al. 2017) and were designed ing association ofces, or public transport providers). to give consideration to a broader range of issues, perti- Exclusion criteria nent to crime prevention practitioners and policymak- •• Workplace violence between colleagues, lateral vio- ers (Johnson et al. 2015). Tis approach to the systematic lence and bullying was excluded as 94% of physical review in practice required a broader extraction of data, assaults are carried out by patients, and family and in addition to the more traditional estimates of efect, to friends of patients (64% by patients and 30% by fam- uncover evidence across these dimensions and to reveal ily and friends) (Spector et al. 2014). Further to this, research and knowledge gaps in the primary studies and it was anticipated that lateral violence may have dif- the evidence base. ferent causal pathways and mechanisms.

Objectives Types of interventions Te systematic review was conducted to evaluate the evi- Personal security alarms (to include mobile and fxed panic dence underpinning the use of personal security alarms buttons, monitored and passive personal alarms), exclud- with the aim of reducing the incidence of assaults against ing medical personal alarms (e.g. fall alarms for elderly). Perkins et al. Crime Sci (2017) 6:11 Page 5 of 19

Types of outcome measures 14. ProQuest International bibliography of the social sci- Primary outcome measures Te primary outcome vari- ences (1951 to 16/08/2016). able was a reduction in physical assaults (recorded or self- 15. ProQuest theses and dissertations (to 16/08/2016). reported), with studies of any duration included. 16. ProQuest Sociological Abstracts (1952 to 16/08/2016). Secondary outcome measures Te secondary out- 17. National Criminal Justice Reference Service come variables included other recorded or self-reported Abstracts Database ProQuest (to 16/08/2016). measures such as non-physical assaults, staf pain, time away from work, depression and low self-esteem, feel- Other sources ings of safety or insecurity, changes in working practices, We searched the following websites and publications for increased confdence or self-efcacy in violence preven- additional reports and other grey literature to cover the tion. same search period:

Other data We have extracted data from studies on 1. Open Grey (http://www.opengrey.eu/) (to March 2016). mechanism, moderators, implementation and economics. 2. Campbell Systematic Reviews (http://www.campbell- collaboration.org/lib/?go=monograph) (to 16/08/2016). Search methods for identifcation of studies 3. NICE (National Institute for Health and Care Excel- Our search methods comprised four parts: frst, we lence) (http://www.evidence.nhs.uk/) (to 16/08/2016). searched electronic bibliographic databases for published 4. TRIP (https://www.tripdatabase.com/index.html) (to work (see below for electronic databases searched); sec- March 2016). ondly, we searched the grey literature for unpublished 5. Google Scholar (to August 2016). work; thirdly, we searched trials registers for ongoing and recently completed trials; fnally, we screened reference We also performed internet searches, using the Google lists of published studies. Te sources searched were cho- search engine, to trace grey literature and organisa- sen based on their coverage of the topic. tions related to prevention of violence to health service staf. Te Ovid MEDLINE(R) search strategy in Appen- Electronic databases dix 1 was adapted as necessary to search all other listed In order to reduce publication and retrieval bias our sources including the internet search. A specialist in search was not restricted by language, date or publica- criminology grey literature at Rutgers (Phyllis Schultze) tion status. We searched the following databases between conducted searches on our behalf. A collaborator at the March and August 2016: College of Policing (Lynn O’Mahony) conducted further searches on our behalf of the National Police Library Cat- 1. Cochrane Library (to March 2016). alogue and POLKA (the Police Online Knowledge Area). 2. CINAHL Plus (EBSCO) (1937 to 16/08/2016). 3. Global Health (OvidSP) (1910 to 2016 Week 32, Data collection and analysis 16/08/2016). Selection of studies 4. ISI WOS: SCI-EXPANDED (1970) & CPCI-S (1990) All studies identifed through the search process were (to 16/08/2016). frst exported to the EndNote bibliographic database 5. Northern Light conference abstracts (2016 week 32, in order to identify and remove duplicate records (this 16/08/2016). commonly occurs where the same record is found in 6. Ovid MEDLINE(R), Ovid MEDLINE(R) In-Process more than one database). Once de-duplication had been & Other Non-Indexed Citations, Ovid MEDLINE(R) carried out, the remaining records were imported into Daily and Ovid OLDMEDLINE(R) (1946 to EPPI-Reviewer 4 software for screening and coding. Tis 16/08/2016). allowed the team to manage coding tasks, assess inter- 7. PROSPERO (to March 2016). rater reliability, and share the results (within the consor- 8. PsycEXTRA (OvidSP) (1908 to March 2016). tium and externally). Two review authors independently 9. PsycINFO (OvidSP) (1806 to 16/08/2016). examined the titles, abstracts, and keywords of electronic 10. PubMed (Present to 18/08/2016). records for eligibility according to the inclusion criteria. 11. Social Policy and Practice (OvidSP) (to 16/08/2016). Results of this initial screening were cross-referenced 12. Applied Social Sciences Index and Abstracts (Pro- between two review authors, and full-texts obtained Quest) (1987 to 16/08/2016). for all potentially relevant reports of studies. Full-texts 13. Criminal Justice Abstracts with Full Text (EBSCO- of potentially eligible studies went through a second- host) (to 16/08/2016). ary screening by each reviewer for fnal inclusion in the Perkins et al. Crime Sci (2017) 6:11 Page 6 of 19

review, with disagreements resolved by discussion. Refer- 5. EMMIE framework. ence lists of all relevant studies were screened for further Efects. eligible studies. Mechanisms. Moderators. Data extraction and management Implementation. Data were coded in EPPI Reviewer, and relevant data Economics. extracted using a standardised data coding set (Appendix 2). Statistical analysis Assessment of risk of bias in included studies Where estimates of efect were available, we intended A review of quality assessment tools for non-randomised to use statistical software (Stata version 14) to conduct studies identifed six tools that were considered to be a meta-analysis. If the included studies were sufciently useful for systematic reviews (Deeks et al. 2003). For this similar (i.e. comparable participants, interventions, and review, we intended to use a modifed framework of one outcomes) we intended to pool the results using a fxed of these tools from the ‘Efective Public Health Practice or random-efects meta-analysis, with standardised mean Project’ (Tomas 2003). We intended to assess the meth- diferences (SMDs) for continuous outcomes and odds odological quality of the study designs and describe each ratios for binary outcomes, and calculate 95% confdence included study against the following criteria: allocation to intervals for each outcome. Te 95% confdence inter- intervention/control; confounders; blinding; data collec- vals indicate a range within which we can be 95% certain tion methods; attrition; fdelity; and follow up. that the true efect lies (Schünemann et al. 2011). A fxed efect meta-analysis would be used if it could be assumed Data synthesis and analysis that each study is estimating the same intervention efect Descriptive analysis We intended to describe all studies (the true efect of intervention is fxed across studies) and that met the inclusion criteria (see Appendix 2 for stand- that diferences in the results are due to chance. A random- ardised data extraction sheet), including: efects meta-analysis would be used to pool efect estimates across studies where they are assumed to vary and 1. Study design. follow some distribution across studies. A random efects Study design; quality. meta-analysis considers the diferent efect estimates as if Data collection methods, modes, and techniques; they were random where there is lack of knowledge about validity of tools. why intervention efects difer between studies (Deeks et al. Statistical and other analyses. 2011). Heterogeneity (variability) among the study efect estimates was to be assessed using a Chi squared test at 2. Participants. a 5% signifcance level and the I-squared (I2) statistic, the Health care occupations. percentage of between-study variability that is due to true High risk occupations. diferences between studies (heterogeneity) rather than due to sampling error. We planned to consider an I2 value 3. Components of intervention. greater than 50% to refect substantial heterogeneity and Country, city. use a random-efects meta-analysis if there was evidence of Setting. heterogeneity. We intended to conduct sensitivity analyses Teoretical basis used in programme design; postu- in order to investigate possible sources of heterogeneity due lated mediators. to study quality. Sensitivity analysis could include repetition Inputs. of the meta-analysis substituting alternative decisions or Comparator (if a controlled evaluation). subgroups (such as adequate vs. inadequate allocation con- cealment; low vs. high attrition) to determine if the fndings 4. Outcomes. are robust in terms of the decision-making process of the Primary outcomes (e.g. reduction in physical meta-analysis used to generate them (Deeks et al. 2011). assaults, recorded or self-reported). In the absence of sufcient homogeneity, we planned to Secondary outcomes (e.g. time away from work, present tables of the quantitative results instead of pooling non-physical assaults). study results in a meta-analysis. Details of each programme were to be presented in a table of study characteristics. Perkins et al. Crime Sci (2017) 6:11 Page 7 of 19

Results were considered to be the most relevant studies available. Results of the search Te full text reports for these four studies were coded in Records from all searches were imported, screened and detail in EPPI Reviewer and are qualitatively described. coded using the EPPI Reviewer 4 software. Initial screen- Te focus of the review then became to determine what ing was shared between three review authors (DB, DCA, might be inferred from the studies available and what CP), who screened the titles, abstracts and keywords of research gaps remain regarding the use of personal 7922 records for potential eligibility according to the alarms. Given the paucity of available studies assessing inclusion criteria. Tis screening resulted in the exclu- the efects of personal alarms, consideration is given to sion of a total of 7710 records. Te review authors iden- how further studies can be devised to aid policy makers tifed a total of 212 records to be assessed for eligibility and practitioners to decide if, when, where and how to using the full text reports. After reviewing the full text of implement the use of personal alarms to reduce violence. all potentially relevant studies, no studies were identifed that met the inclusion criteria, the numbers of records Description of studies on personal alarm use excluded are shown in Fig. 1. Four studies were identifed which provided some measure of associations between personal alarm use and vio- Expanding the review lence (Gale et al. 2002; Gerberich et al. 2005; Landau and Trough the search process, four studies were identi- Bendalak 2008; Farrell et al. 2014). None was based in fed that provided some evidence relating to violence the UK, with one each in New Zealand, USA, Australia and the use of personal security alarms. Tese stud- and Israel. All were based in healthcare settings, with ies did not meet the study design inclusion criteria but two studies reporting on samples of registered nurses

Fig. 1 Flow of studies Perkins et al. Crime Sci (2017) 6:11 Page 8 of 19

(Gerberich et al. 2005; Farrell et al. 2014), one study con- One hundred and eleven respondents reported the ducted on staf in psychiatric units (Gale et al. 2002) and use of preventative measures. Of these 45% provided a further on all personnel in emergency wards (Landau staf with personal alarms and 41% made panic buttons and Bendalak 2008). None of these evaluated the use of available to staf. Study authors compared unit rates for personal alarms in a controlled way, and of these stud- outcome measures to risk/preventative factors using ies, three used a cross sectional design (Gale et al. 2002; the Mann–Whitney U test. Tey reported that the use Farrell et al. 2014; Landau and Bendalak 2008) with the of pocket alarms was associated with a higher level of remaining study (Gerberich et al. 2005) employing a sur- attempted assault (p = 0.0001) and attack (p = 0.0001) vey with a nested case control study design. Response and the use of panic buttons had no signifcant associa- rates for these studies varied from low (30%; Farrell et al. tion with risk of violence. However, the temporal order 2014) to high (78.5%; Landau and Bendalak 2008) with of alarm use and assault risk was not considered in the Gale et al. (2002) and Gerberich et al. (2005) reporting research. moderate response rates (55% and 66–68% respectively). Gerberich et al. (2005) looked at the use of panic but- Gerberich, 2005 tons and personal alarms but grouped the use of personal An initial screening survey was undertaken of 6300 cur- alarms with mobile telephones (distinguishing between rently active nurses in Minnesota who had been work- whether this was owned by the employee or provided by ing for at least 12 months, and were selected randomly the employer). Farrell et al. (2014) did not look individu- from a 1998 licensing database (response rate 78%). A ally at the use of personal alarms either, instead referring nested case control study was conducted using follow to the provision of ‘personal protective equipment’, which up questionnaires to a sub-set of respondents to evalu- was not clearly defned but was said to include personal ate environmental exposures and physical assault. Four duress alarms. Gale et al. (2002) measured the use of both hundred and seventy-fve cases were identifed as nurses personal alarms (referred to as pocket alarms) and panic who reported at least one incident of physical assault buttons, while Landau and Bendalak (2008) looked only within the preceding 12 months. Follow up questions at the availability of panic buttons to staf. Tree of the enquired about conditions during the month when the four studies used self-reported experiences of violence, assault took place (or for the frst event if there had which risk recall and reporting bias (Gerberich et al. been multiple incidents). Responses were received for 2005; Farrell et al. 2014; Landau and Bendalak 2008). Far- 324 cases, a response rate of 68%. Controls were ran- rell et al. (2014), however, asked participants to report domly selected (1425) and also asked about conditions on their last four working weeks to reduce any impact of during researcher-specifed months included in the recall bias and Gerberich et al. (2005) focused on particu- study period before they reported any physical assaults. lar months in part to avoid recall problems. Gale et al. Responses were received from 946 control subjects, a (2002) used violent events reported to psychiatric unit response rate of 66%. managers as their measures of violence but noted that the Multiple logistic regression was used to explore the reporting procedures for events varied between units and relationship between exposures (including environmen- the use of managers may have led to reporting bias. tal and personal protection factors) and physical violence. Exposures were classifed as follows: Gale, 2002 A study in New Zealand used a cross-sectional survey •• General characteristics of staf and their duties— design to determine associations between a number of included work experience (years licensed, years in risk factors and preventative measures and experiences department), average patient contact hours per shift, of violence. An anonymous postal survey was sent to the average number of nurses and staf in immediate managers of all 219 psychiatric units in New Zealand work environment during most commonly worked and completed questionnaires were received from 119 shift; primary area/department worked; main patient (response rate 55%). Questionnaires covered information population; primary activity; on the unit, the number of reported cases of events over •• Environmental design (features of the hospital envi- the last 12 months (outcomes included property damage, ronment)—including lighting, accessibility of exits, attempted assaults, physical attack, sexual harassment physical barriers blocking vision; and stalking) and the provision of preventative measures •• Environmental protection (security features in the (including staf training, use of hazard sheets, pocket hospital environment for staf protection)—included alarms and panic buttons). the presence of video monitors, metal detectors, Perkins et al. Crime Sci (2017) 6:11 Page 9 of 19

security alarm/panic button, controlled access, secu- •• A high standard of patient facilities (to reduce fears, rity personnel, or escort/body guard; and frustration and anxiety; adjusted OR 0.63; 95% CI •• Personal protection (individual protection factors)— 0.48–0.84), included whether or not staf had access to a cellular •• Te provision of personal protective equipment telephone or personal alarm (and whether this was such as mobile phones and personal duress alarms their own or provided by employer) and whether (adjusted OR 0.73; 95% CI 0.56–0.97), staf carried personal protection. •• Sufcient stafng levels (adjusted OR 0.70; 95% CI 0.52–0.92), and Tis study reported: •• Efective policy enforcement by management (adjusted OR 0.56; 95% CI 0.43–0.74). a. Tat of environmental factors, when lighting was less bright than daylight odds of assault were doubled As found in Gerberich et al. (2005) and previously (fully adjusted OR 2.15; 95% CI 1.58–2.83); noted in this review, the healthcare setting was also b. Tat the presence of a security alarm or panic but- found to be critical to PVA occurrence. Farrell et al. ton (vs none) was associated with increased risk of (2014) reported that personnel in accident and emer- assault (fully adjusted OR 1.56; 95% CI 0.96–2.39); gency (Adjusted OR 4.27; 95% CI 2.47–7.39), aged care and (adjusted OR 3.33; 95% CI 2.22–5.00) and psychiatric or c. Tat risk was substantially lower among nurses who mental health departments (adjusted OR 4.09; 95% CI provided their own cellular telephones or portable 2.41–6.95) were signifcantly more likely to be assaulted alarms (fully adjusted OR 0.30; 95% CI 0.15–0.71) but in comparison to other settings. Staf shift patterns were that if these were provided by the employer that there also reported to impact on the likelihood of PVA, with was no diference (fully adjusted OR 1.01; 95% CI higher risk of PVA being attributed to those on rotating 0.70–1.54). Authors suggested the reduced risk was rosters compared to non-rotation (adjusted OR 2.15; 95% not due to the availability of the alarm or telephone. CI 1.58–2.92) and night duty when compared to day duty (adjusted OR 1.86; 95% CI 1.25–2.78). Te study also reported the nurses were at greatest risk when working in nursing home/long-term care/rehabili- Landau, 2008 tation; emergency departments; and psychiatric/behav- Tree thousand self-report questionnaires were dis- ioural departments. Tese fndings accord with those tributed in the emergency wards of Israel’s 25 general reported earlier in this review. hospitals to all personnel (to permanent medical and nonmedical staf, and staf from other wards attending Farrell, 2014 patients in the emergency wards at the time of ques- Questionnaires were sent to a random sample of 5000 tionnaire distribution). Te response rate was 78.5% registered nurses and midwives in Victoria, Australia to (2356 completed questionnaires). A General Exposure explore their experiences of patient and visitor assault to Violence Index (GEVI) was devised, based on par- (PVA). Te response rate was 30%. Te survey asked ticipants’ reports of exposure to violence in the pre- respondents about their experience of any kind of assault ceding 12 months. A multiple regression model was (including verbal, physical and threat of harm) in the pre- used to explain the GEVI, using 15 independent vari- vious 4 weeks, the presence of ‘protective’ factors in the ables, including professional characteristics (e.g. average workplace and their importance in managing workplace weekly hours, position, length of time in department, violence. 36% of respondents reported that they had participation in violence training workshops), personal experienced a patient or visitor assault (PVA) in the pre- characteristics (e.g. country of origin, gender, age) and vious 4 weeks, which included verbal abuse (90% of vic- hospital features (e.g. size and availability of an emer- tims), physical abuse (45% of victims) and threat of harm gency button). (27% of victims). Of these respondents, 46% reported Tis study used routine activity theory (Cohen and repeat victimization (with three or more instances of Felson 1979; Felson et al. 2016) (RAT) as its theoreti- PVA) with patients being responsible for around two and cal framework. RAT’s starting point is that for a crime a half times more incidents of PVA than visitors were. to occur a ‘suitable target’ (in this case a healthcare staf Binary logistic regression analysis was undertaken to member), a ‘motivated ofender’ (such as an angry, frus- identify any association between covariates and outcome trated or mentally unbalanced patient or visitor) and (PVA). Te reported results found a number of ‘protec- absence of a ‘capable guardian’ (one source of which tive factors’ against PVA. Tese included: might be an alarm system) must converge in time and Perkins et al. Crime Sci (2017) 6:11 Page 10 of 19 0.078) on respondents (p = 0.078) on respondents - Unavail violence. to exposure higher risk to of ability related victimisation assault (p = 0.0001) risk of violence Close to signifcant efect efect signifcant Close to 0.73 (0.56–0.97) 1.56 (0.96–2.39) 0.30 (0.15–0.71) 1.01 (0.70–1.54) Associated with increased risk of with increased Associated association with No signifcant Association pared using Mann–Whitney pared U test Multiple regression analysis Multiple regression Adjusted odds ratio and 95% CI Adjusted Adjusted odds ratio and 95% CI Adjusted Unit rates of riskUnit rates factors com - Reported measure Index (GEVI) based on self reported type and frequency of victimisation violence to year during preceding visitor assault, based on self- visitor reported experiences of PVA 4 working most recent over weeks nonphysical violence events violence events nonphysical during the study period (property attempted damage, assault, sexual harassment, stalking) General Exposure to Violence Violence to General Exposure Occurrence of patient and Self-reported and physical Unit reported physical assaults Unit reported physical Outcome (guarding) equipment (such as mobile phones and personal duress alarms) personal alarm alarmtelephone/personal Access to an emergency to button Access Provision of personal protective of personal protective Provision Security alarm/panic button User owned mobile telephone/ User owned mobile provided Employer Pocket alarms Pocket Panic buttons Panic Device - Israel in Victoria, AU vey of Minnesotavey Nurses managers of psychiatric units managers of psychiatric in NZ Survey of personnel in EDs Survey of nurses and midwives Nested case control study, sur study, case control Nested Cross sectional surveyCross of Design Summary of efects in studies found Landau and Bendalak ( 2008 ) 2014 ) ( et al. Farrell Gerberich ( 2005 ) et al. al. ( 2002 ) Gale et al. 1 Table Study Perkins et al. Crime Sci (2017) 6:11 Page 11 of 19

space. If alarms are an efective source of guardianship to reduce assaults against healthcare staf. Studies are the likelihood of victimization will be higher for those at best suggestive but do not allow us to conclude any- without access to an emergency button as a protective thing with regard to efect and causation; they raise only device. possibilities. Te lack of availability of an emergency button was related to an almost statistically signifcant higher risk of Mechanism victimization (p = 0.078). “Respondents without an emer- Personal security alarms may send a deterrent message to gency button, tended to be more victimized” (referring to potential ofenders, while also improving staf confdence the GEVI, based on self reported type and frequency of and decreasing fear. A report from the National Institute victimisation to violence in the preceding year). for Health and Care Excellence (NICE) (2006) suggested that personal security and fxed alarms along with com- Discussion munication devices are a useful way of preventing violent In the discussion, information has been drawn together behaviour and protecting staf where it occurs, but no from the available studies that have been highlighted in research relevant to this was identifed through searches the results section, and additional evidence gathered and undertaken as part of this review. synthesised (from sources including further relevant lit- Te use of a passive (unmonitored) audible alarm is erature and expert opinion, see Appendix 4) to speak primarily intended to disarm and disorientate an ofender to the EMMIE framework, on the efects, mechanisms, giving the victim time to get away. To be efective it is moderators, implementation and economics of personal suggested that the noise emitted needs to be greater alarm use. than 130 dB (ideally around 138 dB) and sounds continu- ously, to avoid similarity to other more frequently heard Efects car alarms (Personal Alarms 2016), although no research It seems apparent that none of the studies found are studies supporting this contention in healthcare settings able to provide clear measurements of efect of personal were uncovered in the course of this review. alarms on assaults against healthcare staf (see Appendix Personal safety alarms are often designed to be dis- 3: Characteristics of studies; Table 1: Summary of efects). creet and hidden either to enable the user to create a At best, these studies provide some evidence of asso- moment of surprise or disorientation in the potential ciation; however the direction of this association difers attacker to allow time to get away, or to call covertly between studies. One study (Gale et al. 2002) found that for physical assistance to de-escalate a situation. While the use of pocket alarms was associated with an increased it has been reported that audible alarms should not be risk of assault (p = 0.0001), while there was no signifcant used in indoor or enclosed environments due to the risk association between panic alarms and risk of violence. of escalating a potentially violent situation, there is no Conversely, three studies (Gerberich et al. 2005; Farrell evidence to establish whether triggering of a silent (cov- et al. 2014; Landau and Bendalak 2008) report that the use ert) or audible personal alarm (overt) can have any sub- of personal protective equipment (mobile phone or per- sequent efect on the outcome in terms of occurrence or sonal alarm) and access to an emergency button may be severity of an assault. Evidence from available studies did associated with some decreased risk of assault. However, not allow us to elicit any fndings on whether a situation Gerberich et al. (2005) found that the association between is more likely to be de-escalated or escalated in particu- personal protective equipment and assault disappeared lar settings as a result of an alarm being triggered overtly when the equipment was provided by employers (rather or covertly. than employee owned), and therefore this association Research conducted into the costs and benefts of mon- cannot be attributed to the protective equipment alone. itored quick-response pendant alarms (linked to police While the searches conducted for this review focused control rooms) to reduce repeat victimization for domes- on intervention based studies, none of the studies intro- tic violence has highlighted a number of preventative duced the use of personal alarms as an intervention. Te mechanisms that may be activated from their use (Farrell studies found do not deal with the temporal order of a et al. 1993). While recommendations around the use of personal alarm intervention and outcome measures and personal alarms in domestic violence situations may have therefore cannot provide evidence of causality. Te stud- changed, these mechanisms describing how personal ies additionally were not conducted in a controlled way alarms may work could also apply to their use in other and lack appropriate comparison groups. Tese stud- contexts, and are described as: ies provide some useful insight into factors that could be further investigated but do not provide evidence 1. Direct incapacitation—response to alarm activation to inform us on the efectiveness of personal alarms prevents assault; Perkins et al. Crime Sci (2017) 6:11 Page 12 of 19

2. Indirect incapacitation—alarm activation leads to will be diluted due to the diferences between settings arrest of ofender preventing further assaults during (e.g. community and ward settings) and it also seems detention; likely that the ways in which alarms might work might 3. Specifc deterrence—potential ofenders are aware of difer. Diferent settings are therefore likely to require dif- personal alarm and as a result do not ofend; ferent alarm systems, and research is needed to identify 4. General deterrence—potential ofenders aware of a those systems with the most potential for success. scheme to equip staf with personal security alarms A related critical condition for personal alarms to pre- and therefore avoid ofending as a result (Farrell et al. vent violence against healthcare staf may concern the 1993). source of the violence, in particular whether assaults are due to unintentional ‘medical reasons’ present at the time In terms of the four studies identifed in this review, of the incident or whether they are criminally motivated. only Landau and Bendalak (2008) suggested a mecha- Tis was only alluded to in the highlighted studies, and nism to explain how personal alarms might act to reduce is intrinsically linked to the healthcare setting and pri- assaults. Te authors developed hypotheses to test the mary patient population. NHS Protect reported that 25% likelihood of victimisation against 15 independent vari- (17,851) of total reported assaults (70,555) in 2015–2016 ables based on routine activity theory. Te authors’ did not involve any medical factors (NHS Protect 2017a). third hypothesis on guarding, suggested that the likeli- Te ambulance sector reported the highest proportion of hood of an individual being victimized will be reduced if assaults determined as having no medical factors at 75% they are exposed to greater levels of guardianship as this with the victim wishing to report the incident to police decreases opportunities for victimization. Te authors in 50.7% of cases. Tis was very diferent in acute and specifcally identifed the access to an emergency button mental health sectors where 26.7 and 21.9% of reported as a protective device that serves to increase guardian- assaults respectively had no medical factors involved and ship. In their specifc setting (emergency department with only 4.9 and 2.8% of assaults with victim wishing to wards in Israel) they found a near statistically signifcant report to police (Dixon 2010). It is reasonable to conjec- association between the lack of access to an emergency ture that any preventive mechanisms of alarm systems button and an increased likelihood of being victimized. will vary by the distribution of sources of violent behav- While this study does not provide a rigorous test for the iour, which might thereby be expected to moderate the suggested mechanism, and does not control for other fac- outcomes produced. tors, the authors do highlight that it demonstrates the Diferent types of alarms may be required to reduce importance of environmental and situational factors in assaults in these diferent scenarios. Personal audible applying interventions to reduce assaults as implied by alarms are unlikely to deter ofenders from committing routine activity theory. assaults when they are not in control of their actions, while they may be more efective in deterring ofenders Moderators in a community setting who have no medical factors. Reported variations in rates of assaults in diferent In circumstances where medical factors contribute to healthcare settings suggest there may be diferent pro- assault, access to a personal or fxed alarm as a compo- cesses and pathways in terms of violence and therefore nent in a monitored system and which trigger immedi- potentially diferent requirements for violence preven- ate assistance may be able to prevent or reduce severity tion in diferent settings. Gerberich et al. (2005) and Far- of assaults. Tere is a range of diferent types of personal rell et al. (2014) both reported higher rates of assault in alarms available as discussed previously in this review elderly care, psychiatric and emergency department and these should be considered to ensure the most settings. NHS Protect note similar variations in rates appropriate technology is implemented for each setting. of assaults across diferent sectors, with fgures for In addition to the diferent clinical settings having an 2015/2016 reporting assault rates per 1000 staf to be 21 impact on the potential for efectiveness of personal for the acute sector, 54 for the ambulance sector and 191 alarms, the studies highlight diverse other underlying for the mental health sector (NHS Protect 2017a). Te factors that may be having a moderating impact at the healthcare setting (e.g. whether community, ambulance, same time. Farrell et al. (2014), noted stafng shift pat- acute, secure wards, in patient or emergency) and patient terns and rotations (higher risk implied for those working population (e.g. geriatric, adult, paediatric or mental nights and in rotating shift patterns), along with the health) have been shown to be strong moderating fac- standard of patient facilities and experience of nursing tors. If preventive intervention eforts are targeted across staf to be a factor in risk of assault. Gerberich et al. whole healthcare sectors, it seems likely that any efects (2005) similarly pointed to ambient lighting being as Perkins et al. Crime Sci (2017) 6:11 Page 13 of 19

bright as daylight as having a signifcant and positive key condition for alarms to maximise their potential pre- efect on the risk of violence. Jason Pott1 (also see Appen- ventive outcome. dix 4) however, noted that depending on who manages However, Jason Pott noted that in his experience in the the hospital (e.g. a private fnance initiative) the Trust NHS, implementation of policies or interventions with may not have control over such environmental factors. good publicity does not always translate well into prac- Where possible, such factors which may be moderators tice on the ground (e.g. the zero tolerance of aggression of the intervention, should also be considered when in the NHS). implementing a scheme. “[zero tolerance] Policies are only partially imple- Implementation mented (i.e. physical aggression, but not verbal aggression). Te zero tolerance message does not Tere is some overlap in the moderating factors and seem to be delivered on the ‘shop foor’; it seems to be issues that should be considered in implementing a per- a ‘headline’ policy and doesn’t deter those individu- sonal alarm intervention but there is little mention of als who are most likely to ofend against NHS staf. the specifcs required to implement such an intervention Tere is also poor implementation and support for successfully, in the studies found. However, in discussion the policy…” with Jason Pott (see Appendix 4), a Lead Research Nurse in a London emergency department, he clearly felt that So, publicity may be a key contextual condition for the practical implementation of any intervention would alarms to produce violence-reducing outcomes but the require careful consideration: practical implementation of publicity is evidently partial and it clearly cannot have its possible moderating infu- “Te Emergency Department is a complex and ence if not put in place. stressed environment where patients are not in con- Pott’s personal experience of implementation of per- trol and are ‘at the mercy of staf’. [Waiting] time is a sonal alarms in an emergency department was also that major factor, in addition to intoxication and pain… they are often in practice patchily operated. For example there are a number of fash points which can trigger he notes that alarms were often misplaced, not handed aggression… Any intervention has to be considered over at shift changes and were not efectively deployed. to terms of its practical application within the ED.” Here, the system might be efective if operated as In arguing that the particular attributes of emergency intended but cannot be if in practice, it is not used. Mon- departments need to be considered in implementing an itored fxed alarms including locational information on alarm scheme if it is to be efective, Pott is suggesting activation might provide the best opportunity to difuse a that these attributes will be important in shaping how potentially assaultive situation. any given alarm system will work through in practice to Te zero tolerance policy to aggression in the NHS has produce its consequences. In other words, the emergency been highly publicised in order to act as a deterrent to department context (and by extension the particular ofenders by increasing the perceived risk of apprehen- attributes of any other healthcare setting) are important sion. Publicising the implementation of alarms could do as implementation factors because they function as mod- the same, although with publicity there is also the poten- erators, conditioning any efects brought about by an tial to increase the fear of crime or violence. A strategic alarm system. approach e.g. to specifcally publicise the success of an A further factor mentioned by Pott but not explicitly intervention is more likely to reassure the public while at considered in any of the highlighted studies, includes the the same time increase the perceived risk of apprehen- use of publicity in deterrence and how this may impact a sion to ofenders (Bowers and Johnson 2005). Careful personal alarm intervention (whether or not if the public implementation of an alarm intervention that considers and potential ofenders are aware of the use alarms this the setting, engages the commitment of staf and man- would increase deterrence). Indeed, there is extensive agement to apply it in practice, and includes appropri- literature on the use of publicity to increase deterrence, ate policy and publicity to support it, could increase the and it has been shown to increase the success of crime deterrence efect. reduction interventions (Laycock 1991; Bowers and Johnson 2003, 2005). Publicity may therefore comprise a Economics Tere was no mention of costs in the four studies high- 1 Involvement of NHS staf was an important part of the development of lighted in this review. Te costs of violence to the NHS, this review. Jason Pott, a Lead Research Nurse in the Emergency Depart- have been noted previously in this review as being ment of the Royal London Hospital, was approached for advice during £69 million per annum for physical and non-physical vio- development of this review as NHS staf involvement was considered important. lence and aggression (National Audit Ofce 2003), with Perkins et al. Crime Sci (2017) 6:11 Page 14 of 19

the cost of physical assaults during 2007–2008 being in the diferent types of settings in which healthcare staf estimated at £60 million (NHS Security Management members are at high risk of assault. Service 2007). Tere are signifcant impacts for the NHS Te underlying theory would specify the preventive of violence against staf, which were mentioned previ- mechanisms that alarms could activate, and the type of ously. Tese include, pain, time away from work, depres- alarm system and complementary conditions that would sion and low self-esteem, which in turn will mean loss of be needed if the preventive mechanisms were to be acti- NHS personnel for signifcant periods of time, high staf vated. In particular, trials to address the diferent catego- turnover and resignations. Violence against staf repre- ries and mechanisms of alarms (audible or silent), and sents substantial costs to the NHS and should therefore their intended outcome (to reduce the severity of assaults justify appropriate levels of investment in order to reduce or deterrence) should be considered. Experimenters its occurrence. would then work closely with those managing and working within the trial sites to operationalize the system in Research needs in the future the theoretically specifed ways. Tey would track both Te risks of assault from patients in particular, but also the capital and maintenance costs involved in the system, from their family and friends, are high and costly in some note implementation hurdles encountered and how they health care settings. It is therefore important the deci- were overcome, and focus on the details of the expected sion-makers have informed guidance on cost-efective short and long-term changes in levels of assault in experi- preventive strategies. mental as against matched comparison sites. Long-term A range of measures can be and have been used to tracking is advocated given the possibility that staf and try to reduce the problem, of which alarms form a patient turnover and adaptation to the alarm system may part. Alarm systems themselves difer widely from one change the way in which the alarm system operates over another. Such alarm systems work in diferent ways and, time. as found where alarms are used as a preventive measure Assuming that the trial produced promising fndings, for other crime problems, they often depend on the avail- subsequent trials could then relax the artifcial conditions ability and inclination of others to respond and take fur- imposed by the efcacy trial to try to discern whether and ther action and those inclinations to respond are apt to how alarms might be applied cost-efectively in packages change over time. Given the diversity of contexts where of measures to reduce violence in healthcare settings. health staf are at risk, the variations in alarm system, and the varying range of other preventive measures in place Conclusion alongside alarms it would have made little sense to aggre- Tere is a clear and urgent need for preventative meas- gate fndings from feld trials. Tere is too much hetero- ures to reduce violence experienced by NHS personnel. geneity and complexity. Tere is evidence of substantial variations in risk across Prior to undertaking further primary studies, a useful diferent healthcare sectors, primary patient popula- starting point for future research would be to undertake tions and clinical conditions. Te literature discusses a wider consultation with healthcare staf, as well as pro- wide range of preventative measures aimed at reducing fessionals from the security industry and organisations assaults. Tere are also diverse types of alarm system specialising in personal safety training, consultancy and available to address the problem. Additionally, diferent confict management, to assess a cross-section of issues assault reduction interventions, including alarms, may that may afect the implementation of any personal alarm interact with one another. In this complex and diverse intervention and ascertain the conditions believed neces- feld, decision makers need to be able to make informed sary for them to produce their intended outcomes. Fol- choices to select measures that will work cost-efectively lowing this, rather than moving straight to feld trials to in their setting. So far, however, little research or evi- meet the obvious evidence gap, it might be prudent frst dence has emerged to estimate their efectiveness, or to to undertake a series of EMMIE informed efcacy tri- address the diferent ways in which diferent systems may als. Efcacy trials are designed to test theory by creat- be efective in varying settings. ing conditions in which measures conjectured to have Tere is a strong need to build a knowledge base that the potential to bring about their intended efects have is ft for purpose and therefore in the short term, the their best chance of doing so (Loudon et al. 2015; God- authors suggest EMMIE-focused efcacy trials to try to win et al. 2003). In contrast feld trials aim to fnd out if gauge the potential preventive benefts from alarms as the measures are efective in practice in real everyday a prelude to future feld trials. Further, if the fndings of conditions. EMMIE informed efcacy trials would expli- the efcacy trials are positive, to ascertain whether their cate the underlying theory of how alarms could produce routine deployment in high-risk settings could be recom- reductions in assaults in the varying conditions present mended as a cost-efective option. Perkins et al. Crime Sci (2017) 6:11 Page 15 of 19

Authors’ contributions 1. Aggression/ The topic for this review was selected in consultation with the College of Policing. All authors, with the exception of NT, contributed to the search 2. violence/or workplace violence/ strategy, data extraction, and analyses. CP and NT drafted the fnal manuscript. 3. Verbal abuse.mp. All authors read and approved the fnal manuscript. 4. Physical abuse.mp. Authors’ information 5. Dangerous Behavior/ Chloe Perkins is a postdoctoral researcher in the Faculty of Epidemiology and 6. anger/or frustration/or hostility/ Population Health at the London School of Hygiene and Tropical Medicine. 7. (language adj3 (menacing or discrimin* or insult* or Her interests include environmental exposure modelling and assessment. Deirdre Beecher is Senior Metadata Specialist with Cochrane. Her interests abuse or threat*)).ab,ti. include search methods, and metadata management and vocabulary 8. (workplace adj3 (disturb* or violence)).ab,ti. building. David Colas Aberg is a Ph.D. student and Research Assistant in the 9. (violen* or aggress* or hostil* or fght* or abuse* or Department of Security and Crime Science, University College London. His research interests include bone mechanics and the diagnoses of accidental accident* or assault*).ab,ti. and non-accidental bone traumas in children. Phil. Edwards is an Associate 10. ((physical or verbal) adj3 (attack* or threat* or abuse* Professor in Epidemiology and Statistics at the London School of Hygiene and or violence or aggression or insult*)).ab,ti. Tropical Medicine. His research interests are in the epidemiology of unintentional injury, particularly injuries caused within the transportation system. He 11. 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8 or 9 or 10 has evaluated the efects on road safety and crime of reduced street lighting 12. ((personal or fxed or portable or emergency) adj3 in the UK. Nick Tilley is a professor in the Department of Security and Crime (alarm* or device*)).ti,ab. Science, University College London. His research interests include evaluation methodology, the international crime drop, situational crime prevention and 13. ((staf or employee*) adj3 (safe* or protect* or policing. secur*)).ti,ab. 14. risk management/or safety management/ or security Author details 1 Faculty of Epidemiology and Population Health, London School of Hygiene measures/ & Tropical Medicine, London, UK. 2 Cochrane Central Executive Team/IKMD, 15. Rape/pc [Prevention & Control] 3 London, UK. Department of Security and Crime Science, University College 16. Sex Ofenses/pc [Prevention & Control] London, 35 Tavistock Square, London WC1H 9EZ, UK. 17. ((panic or alarm or help or SOS) adj3 (bell* or but- Acknowledgements ton*)).ti,ab. The authors would like to thank Rachel Armitage and Saskia Garner for review- 18. ((warning or security or alert*) adj3 device*).ti,ab. ing an earlier version of this paper. Their comments led to substantial improve- ments. This project is part of the University Consortium for Evidence-Based 19. 12 or 13 or 14 or 15 or 16 or 17 or 18 Crime Reduction and was funded by College of Policing and the Economic 20. 11 and 19. and Social Research Council (ESRC); RC Grant Reference: ES/L007223/1. The views and opinions expressed herein are those of the authors and do not necessarily refect those of the College of Policing or ESRC. Notes: [mp.=title, abstract, original title, name of substance Stakeholder involvement word, subject heading word, keyword heading word, pro- Jason Pott, Lead Research Nurse in the Emergency Department of the Royal tocol supplementary concept word, rare disease supple- London Hospital, provided advice to the investigative team in the review development (Appendix 4). In our experience, user input is particularly valu- mentary concept word, unique identifer]. able in considering outcomes of interest to users, and identifying preferred methods of disseminating results to user groups. Appendix 2: Standardised data extraction

Competing interests Text was coded from included studies under the follow- The authors declare that they have no competing interests. ing data extraction categories:

Consent for publication Not applicable. Study design • Meta-analysis; Ethics approval and consent to participate • RCT; Does not apply. As a review of summary data from previously conducted research, no humans (or their tissues) participated as subjects in this research. • Controlled interrupted time series; • Controlled before and after; Funding • Before/after not controlled; The research reported here was funded by the Economic and Social Research Council (ESRC) Grant ES/L007223/1 and the College of Policing. • Cross sectional; • Case study; Appendix 1: Search strategy • Qualitative; Ovid MEDLINE(R), Ovid MEDLINE(R) In-Process & • Commentary. Other Non-Indexed Citations, Ovid MEDLINE(R) Daily and Ovid OLDMEDLINE(R) Perkins et al. Crime Sci (2017) 6:11 Page 16 of 19

Study length Moderators • Dates of before period; Country of implementation • Dates of after period. Setting of intervention (e.g. hospital ED, inpatient department, community setting) Study setting Setting characteristics • Country; Was the intervention implemented in isolation? • Healthcare setting (e.g. community based, emer- (Standalone/package/unclear) gency or inpatient hospital department); What type of technology was used? • Other. Implementation (what is needed to implement the use Study aims of intervention) Implementation factors mentioned Efect Implementation factors stated to be necessary for Methods success Participants Who is responsible for deciding eligibility/location of Intervention (e.g. fxed alarm, personal alarm) intervention sites? Comparator/control Who is responsible for setting up, operating and main- Outcome measures taining the intervention? Quality of evidence. To include description and treat- Obstacles to implementation ment of bias and confounding Resources required Any personnel requirements (training, monitoring of • Matching of intervention and control areas (e.g. the system). comparability of the settings; whether controls are likely to be afected by the intervention); Economics • Blinding of data collection and analyses; Cost of programme set up • Lengths of data collection time period pre- and Annual running costs post-intervention; Cost of a personal alarm • Control of confounders (e.g. was there an assess- Cost of assaults prevented ment of the distribution of confounders between Cost of personnel/hours intervention and control groups?); Direct and indirect benefts • Adjustment for time trends; Beneft cost ratio/cost beneft analysis. • Any other potential sources of bias (regression to the mean); Appendix 3: Characteristics of studies • Selective reporting of results by study authors. Gale, 2002 (New Zealand) Results—where in full text and quantitative results • Diference between groups (include CI); Methods Cross sectional survey questioning • Interpretation. provision of preventative measures and reported violence Mechanism Participants Managers of the 219 psychiatric units in New Zealand Were any of these mechanisms mentioned? Interventions No intervention •• Increasing risk; Outcomes Reported incidents over the last year •• Reducing reward; of: property damage, attempted assault, physical attack, sexual •• Removing excuses; harassment and stalking over last •• Increasing efort; 12 months •• Other; Notes The use of pocket alarms and panic •• Specifc deterrence; buttons were surveyed, and ana- lysed (Mann–Whitney U test) to •• General deterrence. look for associations with reported outcomes Were any mentioned mechanisms tested? Were any intermediate outcome measures taken to assess the activation of the causal mechanism(s) judged to underpin the intervention? Perkins et al. Crime Sci (2017) 6:11 Page 17 of 19

Gerberich, 2005 (USA) Landau, 2008 (Israel)

Methods Survey with nested case–control Methods Cross sectional survey with multi- study, multiple logistic regression variate regression to determine to explore risk factors (environ- contribution of various factors to mental and personal protection explanation of violence against factors) for physical violence emergency ward personnel Participants 6300 Minnesota nurses (selected Participants 3000 questionnaires distributed to randomly from licensing all types of staf in emergency database), nested case control wards of all 25 general hospitals included full questionnaires from in Israel 310 cases and 946 control subjects Interventions No intervention Interventions No intervention Outcomes Self reported violent incidents dur- Outcomes Self-reported cases of assault in the ing the course of their work, in the previous 12 months previous year—used to generate a GEVI (General Index of Exposure Notes Survey included question on the to Violence) use of mobile telephones and personal alarms (either their own Notes Multivariate regression used to or provided by employer) and determine contribution of various occupational violence experience factors (professional characteristics, personal characteristics, hospital structure including avail- Farrell, 2014 (Australia) ability of an emergency button) to violence (GEVI) Methods Cross sectional survey. Binary logistic regression analysis to determine how protective factors were Appendix 4: Stakeholder interview associated with occurrence of PVA Involvement of NHS staf was an important part of the (patient and visitor assault) development of this review. Jason Pott, Lead Research Participants Random sample of nurses and midwives registered with Nurses Nurse in the Emergency Department of the Royal Lon- Board of Victoria don Hospital, provided advice during its development. Interventions No intervention He was selected as a known contact with expertise and Outcomes Prevalence of PVA from self reported experience in the NHS. He told us that violence against assaults, over four most recent staf is multifaceted and he believes a broad approach working weeks PVA defned as “any incident where would be most efective in prevention. an employee is abused, threat- ened or assaulted in circum- “Te Emergency Department is a complex and stances arising out of, or in the stressed environment where patients are not in con- course of their employment, trol and are ‘at the mercy of staf’. [Waiting] time is a where. . .a person to believe that they are in danger of being physi- major factor, in addition to intoxication and pain… cally attacked, and may involve an there are a number of fash points which can trigger actual or implied threat to safety, aggression… Any intervention has to be considered health or wellbeing. . . . Neither intent nor ability to carry out the to terms of its practical application within the ED.” treat is relevant; the key issue is that the behaviour creates a risk to Jason estimates that physical assaults on staf occur health and safety” weekly in the ED at Royal London and verbal assaults Notes Participants responded giving their occur daily. Tere is a pervasive culture that this is the experiences of PVA over their four accepted norm and particularly in ED, while tolerance most recent working weeks also provided information on ‘protec- may be lower in inpatient and outpatient settings. It is a tive factors’ including personal uniquely high stress/risk environment (the Royal Lon- protective equipment (e.g. per- don ED has 14–16 nurses for 450 patients per day, and sonal duress alarms) up to 50 patients in the department at one time). Factors making assaults likely include high emotions and anxiety Perkins et al. Crime Sci (2017) 6:11 Page 18 of 19

(for patients or relatives) and aggressive language is com- Jason told us there is a strong feeling that violence and monly used and accepted. aggression in Emergency Departments remains under- reported. Verbal aggression is almost never reported to “Te system that is implemented needs to be whole the police; and where reported there is often a resistance system…there needs to be agreement from all stake- to arrest anyone, or at least there is inconsistent treat- holders on how all parties will uphold an approach. ment of ofenders. Tere have been accounts of police Inconsistency weakens all the members of the team reluctance to take patients who may have mental/physi- and prevents the public from seeing the zero toler- cal health issues. Tere is also some reluctance from ED ance approach. Red-carding from A&E is almost staf to report anything but the most serious cases to the unheard of, and those who are removed are rarely police, due to the resources required to report and give physically violent, but are chronic patients with statements (i.e. to maintain patient care in a busy ward complex needs which is more concerning.” they cannot aford staf to be away from patients for Jason commented on the range of interventions that 40 min to report an incident). Incidents may be reported have been proposed to reduce violence against NHS staf. but are often not escalated to the police. Reporting needs to be supported at diferent managerial levels, allowing Environmental factors for whether a patient’s ill health is a contributory factor. CCTV this is widely implemented and has its usefulness ‘Zero tolerance’ policies policies are only partially imple- but requires monitoring and is often under resourced. mented (i.e. physical aggression, but not verbal aggres- Tere would be resistance from nursing staf to have sion). Te zero tolerance message does not seem to be these and they could not be used in cubicles where risk is delivered on the ‘shop foor’; it seems to be a ‘headline’ highest, to protect patient privacy. policy and doesn’t deter those individuals who are most Fixed emergency alarms in the past, personal security likely to ofend against NHS staf. Tere is also poor alarms have been issued to ED staf at the Royal London, implementation and support for the policy: the core but these were often forgotten, misplaced, and handover ‘quality’ indicators, upon which EDs are judged, include after each shift was inefcient. Fixed call alarms in cubi- things such as admission/discharge times, times for tri- cles might be a better solution (current ‘arrest’ alarms do age from ambulance, patient satisfaction; while violence not provide location information). against staf is not included. Security guards are well utilised in the Royal London ED. Tey are efective and engage well with staf but are Staf skills anxious about clinical issues and their role in restraint Te NHS provides compulsory 1 day Confict Resolu- of patients. Tis makes them cautious when considering tion training to all front-line staf. Te training includes removing patients. Security staf are only able really to components of self-defence, personal awareness, confict act under instruction of the clinical staf which requires avoidance techniques, ‘proportional force’ and reinforce- resources to manage. ment that staf should not expect to be hit. It costs around Police ofcers police presence in the ED was an efective £250 to send a nurse on the 1 day course. In addition all policy but cannot be resourced. Some clinical staf feel ED staf have ‘de-escalation’ and breakaway training. poorly supported by the police in real situations. Building access most EDs have worked hard to lock down departments and the new emergency department Publisher’s Note at Royal London has restricted access doors. Tis is an Springer Nature remains neutral with regard to jurisdictional claims in pub- ongoing struggle and is impossible to police because lished maps and institutional afliations. patients often arrive with other people and this needs to Received: 30 October 2017 Accepted: 18 November 2017 be tolerated. Ambient environments, lighting and calming décor décor is challenging and if hospitals are managed through a PFI then the Trust does not have control of the decoration. Lighting is always well lit, possibly aggressively so. References Anderson, L., FitzGerald, M., & Luck, L. (2010). An integrative literature review of interventions to reduce violence against emergency department nurses. Practices and policies Journal of Clinical Nursing, 19(17–18), 2520–2530. NHS Security Incident Reporting System (SIRS) is an elec- Bowers, K. J., & Johnson, S. D. (2003). Reducing burglary initiative: The role of publicity in crime prevention. London: Home Ofce. tronic tool to report security incidents on the premises to Bowers, K. J., & Johnson, S. D. (2005). Using publicity for preventive purposes. NHS Protect, to aid in detecting and preventing crime in In N. Tilley (Ed.), Handbook of crime prevention and community safety (pp. a national, regional and sector specifc context. 329–354). Devon, UK: Willan Publishing. Perkins et al. Crime Sci (2017) 6:11 Page 19 of 19

Cohen, L. E., & Felson, M. (1979). Social change and crime rate trends: A routine NHS Protect. (2013). Confict resolution training: Implementing the learning aims activity approach. American Sociological Review, 44(4), 588–608. and outcomes. London: NHS Protect. College of Policing. (2015). Crime Reduction Toolkit: CCTV 2015. Retrieved NHS Protect. (2014). Business plan 2014–2015. London: NHS Protect. from http://whatworks.college.police.uk/toolkit/Pages/Intervention. NHS Protect. (2015). Lone worker estate mapping exercise. aspx?InterventionID 1. NHS Protect. (2017a). Reported physical assaults on NHS staf fgures: NHS Commission HaS. (1997). =Violence and aggression to staf in health services: guid- Business Services Authority. Data on reported physical assaults against ance on assessment and management (2nd ed). Norwich: Commission HaS. NHS staf in England between 2004–2005 and 2015–2016 is available on Deeks, J., Dinnes, J., D’Amico, R., Sowden, A., Sakarovitch, C., & Song, F. (2003). this page. Retrieved February 1, 2017, from https://www.nhsbsa.nhs.uk/ Evaluating non-randomised intervention studies. Health Technology crime-prevention/nhs-protect-statistics. Assessment, 7, 27. NHS Protect. (2017b). A guide for the better protection of lone workers in the Deeks, J. J., Higgins, J. P., & Altman, D. G. (2011). Chapter 9: Analysing data and NHS. undertaking meta-analyses. In J. P. Higgins & S. Green (Eds.), Cochrane NHS Security Management Service. (2010). Cost of violence against NHS staf. handbook for systematic reviews of interventions. Lisa Bero: The Cochrane A report summarising the economic cost to the NHS of violence against Collaboration. staf 2007/2008. Dixon, D. (2016). A fve year analysis of physical assaults against NHS staf in (NICE) NIfHaCE. (2006). Violence: The short-term management of disturbed/vio- England: SIRS/RPA violence report 2010–2015. lent behaviour in in-patient psychiatric settings and emergency departments. Farrell, G., Buck, W., & Pease, K. (1993). The merseyside domestic violence London: NIfHaCE. prevention project: Some costs and benefts. Studies on Crime and Crime Occupational Health & Safety Agency for Healthcare. (2005). Preventing violent Prevention, 2, 21–33. and aggressive behaviour in healthcare: A literature review. Vancouver, Farrell, G. A., Shafei, T., & Chan, S.-P. (2014). Patient and visitor assault on nurses Canada. and midwives: An exploratory study of employer ‘protective’ factors. Oztunc, G. (2006). Examination of incidents of workplace verbal abuse against International Journal of Mental Health Nursing, 23, 88–96. nurses. Journal of Nursing Care Quality, 21(4), 227–232. Farrington, D. P., Gill, M., Waples, S. J., & Argomaniz, J. (2007). The efects of Rew, M., & Ferns, T. (2005). A balanced approach to dealing with violence and closed-circuit television on crime: Meta-analysis of an English national aggression at work. British Journal of Nursing, 14(4), 227–232. quasi multi-site evaluation. Journal of Experimental Criminology, 3, 21–38. Schünemann, H. J., Oxman, A. D., Vist, G. E., Deeks, J. J., Glasziou, P., & Guyatt, G. Felson, M., & Eckert, M. (2016). Crime and Everday Life (5th ed.). Uitgever: SAGE H. (2011). Chapter 12: Interpreting results and drawing conclusions. In J. Publications, Inc. P. Higgins & S. Green (Eds.), Cochrane handbook for systematic reviews of Gale, C., Pellett, O., Coverdale, J., & Paton, Simpson G. (2002). Management interventions (version 5.1.0 ed.). Lisa Bero: The Cochrane Collaboration. of violence in the workplace: A New Zealand survey. Acta Psychiatrica Sherman, L. (2013). The rise of evidence-based policing: Targeting testing and Scandinavica, 412, 41–43. tracking. Crime and Justice, 42(1), 337–451. Gerberich, S. G., Church, T. R., McGovern, P. M., Hansen, H., Nachreiner, N. M., Sidebottom, A., Tompson, L., Thornton, A., Bullock, K., Tilley, N., Bowers, K. J., Geisser, M. S., et al. (2005). Risk factors for work-related assaults on nurses. et al. (2017). Gating alleys to reduce crime: A meta-analysis and realist Epidemiology, 16(5), 704–709. synthesis. Justice Quarterly. http://dx.doi.org/10.1080/07418825.2017.12 Godwin, M., Ruhland, L., Casson, I., MacDonald, S., Delva, D., Birtwhistle, R., et al. 93135 (2003). Pragmatic controlled clinical trials in primary care: The struggle Spector, P., Zhou, Z., & Che, X. (2014). Nurse exposure to physical and non- between external and internal validity. BMC Medical Research Methodol- physical violence, bullying and sexual harassment: A quantitative review. ogy, 3(1), 28. International Journal of Nursing Studies, 51, 72–84. Health and Safety Executive. (2017a). Work-related violence: Lone workers: Suzy Lamplugh Trust. (2016). Personal alarms 2016. Retrieved February 12, Health and Safety Executive. Retrieved June 20, 2017, from http://www. 2016. hse.gov.uk/violence/conclusions.htm#a2. Thomas, H. (2003). Quality assessment tool for quantitative studies. Toronto: Health and Safety Executive. (2017b). Violence at work. Health and Safety McMaster University. Executive. Wells, J., & Bowers, L. (2002). How prevalent is violence towards nurses work- Home Ofce. (2017). Research Study 217: The economic and social costs of crime. ing in general hospitals in the UK? Journal of Advanced Nursing, 39(3), London: Home Ofce. 230–240. Ipsos MORI. (2010). Violence against frontline NHS staf—research for COI on Welsh, B. C., & Farrington, D. P. (2009). Public area CCTV and crime prevention: behalf of the NHS Security Management Service. London: Ipsos MORI. An updated systematic review and meta-analysis. Justice Quarterly, 26(4), Johnson, S. D., Tilley, N., & Bowers, K. J. (2015). Introducing EMMIE: An evidence 716–745. rating scale to encourage mixed-method crime prevention synthesis reviews. Journal of Experimental Criminology, 11(3), 459–473. Landau, S. F., & Bendalak, Y. (2008). Personnel exposure to violence in hospital emergency wards: A routine activity approach. Aggressive Behavior, 34(1), 88–103. Laycock, G. (1991). Operation identifcation or the power of publicity? Security Journal, 2, 67–71. Loudon, K., Treweek, S., Sullivan, F., Donnan, P., Thorpe, K. E., & Zwarenstein, M. (2015). The PRECIS-2 tool: designing trials that are ft for purpose. British Medical Journal, 350, 2147. National Audit Ofce. (2003). A safer place to work. Protecting NHS Hospital and ambulance staf from violence and aggression. London: Ofce NA. NHS Counter Fraud and Security Management Service. (2003). Protecting your NHS—a professional approach to managing security in the NHS. London: NHS.