Which characteristics should a nurse observe a client for that may be predictive of aggression and violence?

4.1. INTRODUCTION

The identification and management of risk for future violence has become an increasingly important component of psychiatric practice. The Royal College of Psychiatrists, for example, emphasises its commitment ‘to minimising risk in psychiatric practice’ and describes risk management as ‘the guiding force behind all recent reports’ of the College (Morgan, 2007) while also recognising that risk cannot be eliminated. In the UK, conducting risk assessments on psychiatric patients has become part of routine practice in general adult psychiatric settings and most NHS Trusts mandate the use of specific tools. Nevertheless, early data has shown that only about 60% of patients were actually risk assessed (Higgins et al., 2005). It is likely that this figure has since risen, but no recent audit data is available. In forensic settings, national guidance requires high and medium secure service providers to conduct a HCR-20 (History – Risk – Clinical) on all patients. Again, no data is available regarding the compliance with this requirement, although given the inclusion of risk assessment in Commissioning for Quality and Innovation targets in these settings completion rates are likely to be high.

Despite this widespread implementation of risk assessment, driven largely by public concern, it remains uncertain which factors are associated with violence and how to best assess risk. While consensus exists that structured risk assessment is superior to ‘unaided clinical judgement’ alone, a number of recent reviews on risk assessment instruments, such as Fazel and colleagues (2012) and Yang and colleagues (2010), have found their predictive validity to be modest at best and have concluded that the current evidence does not support sole reliance on such tools for decision-making on detention or release of individuals with mental health problems. To complicate matters further, risk assessment is not just a scientific or clinical endeavour, but carries a significant political dimension – which level of risk is acceptable (even if it can be identified accurately) and how to weigh the consequences of false positive and false negative (when it is predicted that violent and aggressive behaviour will not occur, but it does) assessments is ultimately for society as a whole to decide.

4.2. REVIEW PROTOCOL

The review protocol summary, including the review questions and the eligibility criteria used for this chapter, can be found in Table 7 (risk factors) and Table 8 (prediction instruments). A complete list of review questions can be found in Appendix 5; information about the search strategy can be found in Appendix 10; the full review protocols can be found in Appendix 9).

Table 7

Clinical review protocol summary for the review of risk factors.

Table 8

Clinical review protocol summary for the review of prediction.

The review of risk factors was restricted to prospective cohort studies that used multivariate models to look for independent risk factors. The review strategy primarily involved a meta-analysis of odds ratios for the risk of violence for each risk factor or antecedent. Additionally, results from studies that examined the correlation between multiple factors and violence (reported as R2 or Beta) are presented alongside the meta-analysis. Studies only presenting data from univariate analyses (unadjusted results) were excluded from the review.

The review of predictive instruments included prospective or retrospective cross-sectional/cohort studies which presented outcomes that could be used to determine sensitivity and specificity. Additionally, sensitivity and specificity were plotted using a summary receiver operator characteristic (ROC) curve.

4.3. RISK FACTORS FOR VIOLENCE AND AGGRESSION

4.3.1. Introduction

Risk, according to the Oxford Dictionary of English, can be defined as ‘a situation involving exposure to danger’. It is the probability of an uncertain outcome occurring caused by a combination of factors (risk factors) that – if known – offer a chance to intervene to prevent the outcome from happening. In addition to the likelihood of the negative event occurring, how soon it is likely to occur and the expected severity of the outcome are important considerations.

In the context of this guideline, risk factors are characteristics of service users (or their environment and care) that are associated with an increased likelihood of that individual acting violently and/or aggressively. These risk factors can be divided into static and dynamic factors (Douglas & Skeem, 2005). Static risk factors are historical and do not change, such as family background, childhood abuse or seriousness of offending. Age and gender also fall within this category. Dynamic risk factors, on the other hand, are changeable and hence offer the opportunity for intervention. Examples include current symptoms, use of alcohol or illicit substances and compliance with treatment. Risk assessment involves the identification of risk factors and an estimation of the likelihood and nature of a negative outcome while risk management puts in place strategies to prevent these negative outcomes from occurring or to minimise their impact. Some authors have argued that static factors may be better for long-term predictions while dynamic factors may be more suited for the assessment of violence risk in the short term (Douglas & Skeem, 2005).

A large body of literature exists on risk factors for violence, including in individuals with mental disorders (Bo et al., 2011; Cornaggia et al., 2011; Dack et al., 2013; Papadopoulos et al., 2012; Reagu et al., 2013; Witt et al., 2013). The largest of these (Witt et al., 2013) was a systematic review and meta-analysis of risk factors in people with psychosis, providing data from 110 studies and over 45,000 individuals. The authors found that 146 risk factors had been examined in these studies. In line with findings from other studies, criminal history was found to be the strongest static risk factor. Dynamic factors included hostile behaviour, impulsivity, recent drug or alcohol misuse, ‘positive symptoms’ of psychosis and non-adherence with therapy (including psychological and medication). While the factors identified by Witt and colleagues (2013) are based on a large body of evidence, it is of note that considerable heterogeneity exists in the samples studied with regards to the nature of the violence, the way in which the outcome was measured and the clinical settings involved.

Current practice

Failings in the care provided to mentally ill individuals have been highlighted by a number of high profile cases of mentally ill patients committing serious acts of violence and subsequent inquiries into their care in the 1990s2. Since then, mental health practise in the UK has seen an increased focus on risk and guidance has been produced to aid the process of risk assessment and management (Department of Health, 2007; Royal College of Psychiatrists, 2007). These documents stipulate that each patient's risk should be routinely assessed and identify a number of best practice recommendations.

The Department of Health best practice guidance outlines the following as key principles in risk assessment: awareness of the research evidence, positive risk management, collaboration with the service user, recognising their strengths, multidisciplinary working, record keeping, regular training and organisational support of individual practitioners. It further emphasises the importance of ‘risk formulation’; that is, a process that ‘identifies and describes predisposing, precipitating, perpetuating and protective factors, and how these interact to produce risk’ (Department of Health, 2007). This formulation should be discussed with the service user and a plan of action produced as to how to manage the risks identified. Tool-based assessments (as outlined below) should form part of a thorough and systematic overall clinical assessment. It is suggested that given the fluidity of risk, its assessment should not be a one-off activity but should be embedded in everyday practice and reviewed regularly.

Definition of risk factors and antecedents for predicting violence

For the purposes of this review, risk factors and antecedents were categorised using the psychosocial and clinical domains described by Witt and colleagues (2013):

1. Demographic and premorbid

2. Criminal history

3. Psychopathological, positive symptoms and negative symptoms

4. Substance misuse

5. Treatment-related

6. Suicidality.

4.3.2. Studies considered3

For the review of risk factors (see Table 7 for the review protocol), 13 studies (N = 5380) met the eligibility criteria: Amore 2008 (Amore et al., 2008), Chang 2004 (Chang & Lee, 2004), Cheung 1996 (Cheung et al., 1996), Ehmann 2001 (Ehmann et al., 2001), Hodgins 2011 (Hodgins & Riaz, 2011), Kay 1988 (Kay et al., 1988), Ketelsen 2007 (Ketelsen et al., 2007), Kho 1998 (Kho et al., 1998), Oulis 1996 (Oulis et al., 1996), Palmstierna 1990 (Palmstierna & Wistedt, 1990), UK700 (Dean et al., 2006; Thomas et al., 2005), Watts 2003 (Watts et al., 2003) and Yesavage 1984 (Yesavage, 1984). Of these, all 13 were published in peer-reviewed journals between 1984 and 2011. In addition, 528 studies failed to meet eligibility criteria for the guideline. Further information about both included and excluded studies can be found in Appendix 13.

Of the 13 eligible studies, 7 (N = 3903) included sufficient data to be included in the statistical analysis. Of those, 5 involved adult participants in an inpatient setting and 2 involved adult participants in a community setting. Table 9 contains a summary of the study characteristics of these studies. Of the 6 studies not included in the analysis, 3 (Ehmann 2001, Kay 1988, Kho 1998) reported no usable data, and 3 (Oulis 1996, Palmstierna 1990, Yesavage 1984) reported statistics that made synthesis with the other studies very difficult. However, the latter 3 studies used very small samples (ranging from 70 to 136) and therefore the results from these studies are not included here as it was felt they would not be useful for making recommendations.

Table 9

Summary of study characteristics for the review of risk factors for violence and aggression in adults.

4.3.3. Evidence for risk factors in adults

All studies reported below had generally low risk of bias, except for the domain ‘loss to follow-up’, which was often unclear due to non-reporting (see Appendix 11 for further information).

Demographic and premorbid factors

As can be seen in Table 10, which shows the demographic and premorbid factors in the multivariate model for each study, only 2 factors (age and gender) were commonly included.

Table 10

Demographic and premorbid factors included in the multivariate model for each study.

Age

In 5 studies of 2944 adults in inpatient settings (Amore 2008, Chang 2004, Cheung 1996, Ketelsen 2007, Watts 2003), there was evidence that age was unlikely to be associated with the risk of violence and/or aggression on the ward.

In 2 studies of 1031 adults in community settings (Hodgins 2011, UK700), there was evidence that was inconsistent as to whether age was associated with the risk of violence in the community.

Gender

In both inpatient (Amore 2008, Chang 2004, Cheung 1996) (N = 634) and community (Hodgins 2011, UK700) (N = 1031) settings, the evidence was inconclusive as to whether male gender was associated with the risk of violence.

Ethnicity

In 1 study of 100 adults in an inpatient setting (Watts 2003), there was evidence that African ethnicity was associated with a reduced risk of violence, but the evidence was inconclusive as to whether African–Caribbean ethnicity was associated with a reduced risk.

In 1 study of 780 adults in community settings (UK700), there was evidence that non-white ethnicity was associated with an increased risk of violence. In a sub-sample of 304 women, there was evidence that African–Caribbean ethnicity was associated with an increased risk of violence in the community.

Living in supported housing

In 1 study of 2210 adults in an inpatient setting (Ketelsen 2007), there was evidence that previous residence in supported accommodation was associated with an increased risk of violence and/or aggression on the ward.

In 1 study of 780 adults in the community (UK700), there was inconclusive evidence as to the association between previous residence in supported accommodation and the risk of violence in the community.

Other demographic and premorbid factors

In 1 study of 780 adults in community settings (UK700), there was evidence that history of being victimised was associated with an increased risk of violence but the association was inconclusive for history of homelessness, marital status and past special education. In a sub-sample of 304 women, there was evidence that unmet needs and history of being victimised were associated with an increased risk of violence in the community.

Criminal history factors

In the inpatient setting, no criminal history factors were included in more than 1 study, and in the community setting, only 1 factor (lifetime history of violence) was included in both studies (Table 11).

Table 11

Criminal history factors included in the multivariate model for each study.

Conduct disorder

In 1 study of 251 adults in the community (Hodgins 2011), there was inconclusive evidence regarding whether the presence of a conduct disorder was associated with an increased risk of violence in the community.

History of aggression

In inpatient settings, in 1 study of 303 adults (Amore 2008) there was evidence that recent (past month) and lifetime history of physical aggression and recent verbal or against object aggression were associated with an increased risk of violence on the ward. However, the evidence was inconclusive as to whether a history (lifetime) of verbal or against object aggression was associated with the risk of violence. In 1 study of 100 inpatients (Watts 2003), there was evidence that violence in the 24 hours prior to admission was unlikely to be associated with violence on the ward.

In 1 study of 780 adults in community settings (UK700), there was evidence that a history of physical aggression was associated with increased risk of violence, and in the subsample of 304 women, there was evidence that a conviction for non-violent offense was associated with an increased risk of violence in the community.

Psychopathological, positive symptom and negative symptom factors

In the inpatient setting only 2 factors (diagnosis of a mood disorder and hostility-suspiciousness) were included in more than 1 study, and in the community setting only 1 factor (number of threat/control-override delusions) was included in both studies (Table 12).

Table 12

Psychopathological, positive symptom and negative symptom factors included in the multivariate model for each study.

Onset of psychotic disorder

In 1 study of 111 adults in inpatient wards (Chang 2004), there was evidence that later onset of a psychotic disorder was associated with an increased risk of violence on the ward.

Diagnosis

In 1 study of 2210 adults in inpatient wards (Ketelsen 2007), there was evidence that presence of schizophrenia was associated with an increased risk of violence and/or aggression on the ward.

In 1 study of 303 adult inpatients (Amore 2008), there was inconclusive evidence as to whether a mood disorder (anxiety or depression) was associated with an increased risk of violence on the ward.

In 1 study of 251 adults in community settings (Hodgins 2011), there was inconclusive evidence as to whether the presence of anxiety was associated with an increased risk of violence in the community.

Other symptoms

In 2 studies of 403 adults in inpatient settings (Amore 2008, Watts 2003), 1 study was inconclusive, but the other found evidence that hostility-suspiciousness was associated with an increased risk of violence on the ward. In 1 study of 303 adults in inpatient wards (Amore 2008), there was inconclusive evidence as to whether a thought disturbance, the presence of tension or excitement or lethargy were associated with an increased risk of violence.

In 1 study of 780 adults in the community (UK700), there was evidence that the presence of a personality disorder was associated with an increased risk of violence, and in 2 studies of 1031 adults in the community (Hodgins 2011, UK700) there was evidence that the presence of threat/control-override delusions was associated with an increased risk of violence.

Treatment-related factors

In the inpatient setting, only 2 factors (duration of hospitalisation and number of previous admissions) were included in more than 1 study, and in the community setting, no factors were included in both studies (Table 13).

Table 13

Treatment-related factors included in the multivariate model for each study.

Duration of hospitalisation

In 2 studies of 331 adult inpatients (Chang 2004, Cheung 1996), there was evidence that duration of hospitalisation was not associated with an increased risk of violence on the ward.

In 1 study of 780 adults in the community (UK700), there was inconclusive evidence as to whether longer duration of hospitalisation was associated with an increased risk of violence in the community.

Referral route and admission

In 1 study of 2210 adult inpatients (Ketelsen 2007), there was evidence that referral by a crisis intervention team, home staff (for service users who live in supported housing), and involuntary admission were associated with an increased risk of violence and/or aggression. In addition, higher number of previous admissions and younger age at first admission were associated with a very small increased risk of violence and/or aggression. In contrast, referral by the doctor with regular responsibility for the service user was associated with a reduced risk.

Substance misuse factors

In the inpatient setting, no substance misuse factors were included, and in the community setting, recent drug use was the only factor and this was included in both studies (Table 14).

Table 14

Substance misuse factors included in the multivariate model for each study.

Previous drug use

In 2 studies of 1031 adults in community settings (Hodgins 2011, UK700), there was evidence that indicated an association between recent (past 6 or 12 months) drug use and the risk of violence in the community.

Suicidality factors

In the inpatient setting, no suicidality factors were included, and in the community setting, previous attempted suicide was the only factor and this was included in only 1 study (Table 15).

Table 15

Suicidality factors included in the multivariate model for each study.

Previous attempted suicide

One study of 780 adults in the community (UK700) examined previous attempted suicide as a potential risk factor for violence, but the evidence was inconclusive.

4.3.4. Health economic evidence

Identification of risk factors for violent and aggressive behaviour by mental health service users in health and community care settings may lead to better prediction of incidents of violence and aggression and has therefore potentially important resource implications. However, this review question is not relevant for economic analysis.

4.4. PREDICTION AND ANTICIPATION OF VIOLENCE AND AGGRESSION

4.4.1. Introduction

Prediction is the cornerstone of the assessment, mitigation and management of violence and aggression. The prediction of violence and aggression is challenging due to the diversity of clinical presentation and it is unlikely that a single broad predictive (assessment) tool could be valid and reliable in all circumstances where violence and aggression needs to be predicted. This is not surprising given that the prevalence of violence and aggression varies considerably in different clinical settings; the prevalence would vary markedly between the community, an inpatient psychiatric ward and a forensic setting. Furthermore, the baseline prevalence of what one is trying to predict is important when considering the utility of the prediction tool.

Fundamentally, the process of prediction requires 2 separate assessments. The application of the prediction tool constitutes the first assessment, and categorises the patient into a lower or higher risk of exhibiting the future behaviour one is interested in predicting. Further down the line, the second assessment concludes whether the patient did or did not exhibit the behaviour of interest. As an instrument, the prediction tool's statistical properties are relevant in assessing its clinical utility. False positives (when the prediction tool identifies that violence and aggression will occur, but it does not) are especially troublesome in this respect, as they can lead to unnecessarily restrictive clinical interventions for the patient. False negatives (when the prediction tool identifies that violence and aggression will not occur, but it does) can have serious consequences for the patient, clinicians and potential victims of the violence or aggression. In reality there is a balance between true and false predictions, which needs to be equated with the consequences thereof.

Translating this process into the clinical or research setting is difficult. The majority of violence and aggression risk assessment tools (prediction tools) are not designed to be completed in minutes to allow for rapid screening, and, if they are designed to be completed expeditiously, they often incorporate a phase of retrospective monitoring of behaviour. The behaviour of interest is violence and aggression, and there is a complex and often unclear relationship between the variables in risk assessment tools, the process of conducting a risk assessment, and the occurrence further down the line, of violence and aggression. An interesting example in this area is the idea that the mere process of conducting a risk assessment may change the probability of future violence and aggression, by either better structuring the ongoing clinical care of the patient or by changing their clinical pathway (for example, to a more secure clinical setting) (Abderhalden et al., 2004).

With such obstacles to prediction of violence and aggression, the question is raised of whether accurate prediction is even possible. Yet in mental health and criminal justice settings, and increasingly in the wider health and social care setting, there is anecdotal evidence that violence and aggression is a major factor inhibiting the delivery of effective modern day services. Currently there is a genuine drive to achieve parity between mental and physical healthcare for patients in the health and social care system. Given that violence and aggression is often associated with a clinical psychiatric emergency, 1 way to raise the profile of the management of violence and aggression may be to consider it to be on a par with more classical medical and surgical emergencies that clinicians encounter in the general hospital setting.

In inpatient psychiatric settings, early detection and intervention with people at risk of behaving aggressively is crucial because once the aggression escalates, nurses are left with fewer and more coercive interventions such as sedation, restraint and seclusion (Abderhalden et al., 2004; Gaskin et al., 2007; Griffith et al., 2013; Rippon, 2000). In this sense, early detection has implications for a more therapeutic and safer patient and staff experience.

Clinical experience and research has led to a plethora of identified violence and aggression risk variables (static, dynamic, patient-related, environmental), which provide the predictive input for risk assessment tools. The utility of predictive risk assessment tools can only be as good as the robustness of the violence and aggression risk variables. In this guideline, the focus is on the evaluation of predictive risk assessment tools and their utility in the prediction of imminent violence and aggression.

Definition and aim of intervention

Prediction instruments (actuarial and structured clinical judgement) can be used to assign service users to 2 groups: those predicted to become violent or aggressive in the short-term and those predicted not to become violent or aggressive in the short-term. In this context, an actuarial assessment is a formal method to make this prediction based on an equation, a formula, a graph, or an actuarial table. Structured professional and clinical judgement involves the rating of specified risk factors that are well operationalised so their applicability can be coded reliably based on interview or other records. Based on this, clinical judgement is used to come to a decision about risk, rather than using an established algorithm (Heilbrun et al., 2010). In addition, the risk factors included in a prediction instrument can be static or dynamic (changeable), and it is the latter that are thought to be important in predicting violence in the short-term (Chu et al., 2013).

There is a long history of research demonstrating that unaided clinical prediction is not as accurate as structured or actuarial assessment (Heilbrun et al., 2010), therefore unstructured clinical judgement is not included in this review.

For the purposes of the guideline, prediction instruments were defined as checklists of service user characteristics and/or clinical history used by members of staff to predict imminent violent or aggressive behaviour (commonly in the next 24 hours).

The behaviour being predicted could range from verbal threats to acts of aggression directed at objects or property to physical violence against other service users or staff.

Methodological approach

When evaluating prediction instruments, the following criteria were used to decide whether an instrument was eligible for inclusion in the review:

• Primary aim of the instrument: the prediction of imminent violence and aggression.

• Clinical utility: the criterion required the primary use of the prediction instrument to be feasible and implementable in a routine clinical care. The instrument should contribute to the identification of further assessment needs and therefore be potentially useful for care planning.

• Tool characteristics and administrative properties: the prediction instrument should have validated cut-offs in the population of interest. Furthermore, and dependent on the practitioner skill set and the setting, instruments were evaluated for the time needed to administer and score them as well as the nature of the training (if any) required for administration or scoring. An instrument should be easy to administer and score, and be able to be interpreted without extensive and specialist training.

• Population: the population being assessed reflects the scope of this guideline. The instrument should have been validated in adults and/or children and young people and preferably be applicable to the UK, for example by being validated in a UK population or a population that is similar to UK demographics.

• Psychometric data: the instrument should have established reliability and validity. In addition, it should have been tested against a gold standard assessment of violence and aggression (direct observation and recording of events), for which sensitivity and specificity is either reported or can be calculated. The sensitivity of an instrument refers to the probability that it will produce a true positive result when given to a population with the target disorder (as compared with a reference or ‘gold standard’). The specificity of an instrument refers to the probability that a test will produce a true negative result when given to a population without the target disorder (as determined by a reference or ‘gold standard’). When evaluating the sensitivity and specificity of the different instruments, the GDG examined both in tandem and used the following definitions as a general rule-of-thumb: values above 0.9 were defined as ‘excellent’, 0.8 to 0.9 as ‘good’, 0.5 to 0.7 as ‘moderate’, 0.3 to 0.4 as ‘low’ and less than 0.3 as ‘poor’.

The qualities of a particular tool can be summarised in an ROC curve, which plots sensitivity (expressed as a proportion) against (1-specificity). Finally, positive (LR+) and negative (LR-) likelihood ratios are thought not to be dependent on prevalence. LR+ is calculated by sensitivity/(1-specificity) and LR- is (1-sensitivity)/specificity. A value of LR+ >5 and LR- <0.3 suggests the test is relatively accurate (Fischer et al., 2003).

See Chapter 3 for further information about the methodology used for this review.

4.4.2. Studies considered4

For the review of prediction instruments (see Table 8 for the review protocol), 10 studies (N = 1659) met the eligibility criteria: Abderhalden 2004 (Abderhalden et al., 2004), Abderhalden 2006 (Abderhalden et al., 2006), Almvik 2000 (Almvik et al., 2000), Barry-Walsh 2009 (Barry-Walsh et al., 2009), Chu 2013a (Chu et al., 2013), Griffith 2013 (Griffith et al., 2013), McNiel 2000 (McNiel et al., 2000), Ogloff 2006 (Ogloff & Daffern, 2006), Vojt 2010 (Vojt et al., 2010), Yao 2014 (Yao et al., 2014). All were published in peer-reviewed journals between 2000 and 2014. In addition, 528 studies failed to meet eligibility criteria for the guideline. Further information about both included and excluded studies can be found in Appendix 13.

Of the 10 eligible studies, 6 (Abderhalden 2004, Abderhalden 2006, Almvik 2000, Chu 2013a, McNiel 2000, Yao 2014) included sufficient data to be included as evidence. As the reference standard, 3 studies (Abderhalden 2004, Abderhalden 2006, Almvik 2000) used the SOAS-R or a modification of this to record all violent and aggressive incidents in the shift following the index test. Two studies (Chu 2013a, McNiel 2000) used the OAS, and violence data and preventive measures were concurrently collected from nursing records and case reports by 1 study (Yao 2014).

4.4.3. Prediction instruments included in the review

Data were available for 2 actuarial prediction instruments: the BVC (Almvik & Woods, 1998) and the DASA – Inpatient Version (DASA-IV) (Ogloff & Daffern, 2002). In addition, the Clinical Scale from the HCR-20 (Webster et al., 1997) structured clinical judgment instrument was assessed in 1 study. See Table 16 for further information about each instrument.

Table 16

Summary of characteristics for each included prediction instrument.

4.4.4. Evidence for prediction instruments

All studies reported below had generally a low risk of bias except for the domain covering the reference standard, which was assessed by staff who also completed the instrument being investigated (see Appendix 11 for further information).

In 4 studies of 679 adults in an inpatient or forensic setting, the BVC using a cut-off of ≥2 had a pooled sensitivity of 0.71 (95% CI, 0.61 to 0.80) and specificity of 0.89 (95% CI, 0.87 to 0.91), and AUC (area under the curve) = 0.93; pooled LR+ = 7.71 (95% CI, 6.20 to 9.59), I2 = 0%; pooled LR- = 0.32 (95% CI, 0.24 to 0.44), I2 = 0%.

In 4 studies of 870 adults in an inpatient or forensic setting, the BVC using a cut-off of ≥3 had a pooled sensitivity of 0.60 (95% CI, 0.52 to 0.67) and specificity of 0.93 (95% CI, 0.92 to 0.94) and AUC = 0.85; pooled LR+ = 8.74 (95% CI, 7.25 to 10.53), I2 = 0%; pooled LR- = 0.44 (95% CI, 0.37 to 0.53), I2 = 0%.

In 1 study of 300 adults in an inpatient setting, the BVC combined with a visual analogue scale using a cut-off of ≥7 had a sensitivity of 0.68 (95% CI, 0.59 to 0.76) and specificity of 0.95 (95% CI, 0.94 to 0.96).

In 1 study of 300 adults in an inpatient setting, the DASA using a cut-off of ≥2 had a sensitivity of 0.88 (95% CI, 0.62 to 0.98) and specificity of 0.59 (95% CI, 0.45 to 0.72) and LR+ = 2.15; LR- = 0.21.

In 1 study of 300 adults in an inpatient setting, the DASA using a cut-off of ≥3 had a sensitivity of 0.81 (95% CI, 0.54 to 0.96) and specificity of 0.69 (95% CI, 0.54 to 0.80) and LR+ = 2.58; LR- = 0.27.

In 1 study of 70 adults in a forensic setting, the HCR-20 Clinical Scale using a cut-off of ≥3 had a sensitivity of 0.88 (95% CI, 0.62 to 0.98) and specificity of 0.41 (95% CI, 0.28 to 0.55) and LR+ = 1.48; LR- = 0.31.

In 1 study of 70 adults in a forensic setting, the HCR-20 Clinical Scale using a cut-off of ≥4 had a sensitivity of 0.81 (95% CI, 0.54 to 0.96) and specificity of 0.52 (95% CI, 0.38 to 0.66) and LR+ = 1.69; LR- = 0.36.

For comparison, 1 study of 470 adults in an inpatient setting that evaluated unstructured clinical judgement is included here. When doctors and nurses independently agreed about the risk, the sensitivity was 0.17 (95% CI, 0.09 to 0.29) and specificity was 0.99 (95% CI, 0.97 to 0.99), and LR+ = 11.86; LR- = 0.84. When doctors and nurses did not agree, the sensitivity was 0.31 (95% CI, 0.20 to 0.44) and specificity was 0.93 (95% CI, 0.90 to 0.95), and LR+ = 4.62; LR- = 0.74.

Figure 1

Forest plot of sensitivity and specificity for instruments used to predict violence in the short-term.

Figure 2

Summary ROC curve for the prediction of violence in the short-term.

Figure 3

Forest plots of pooled sensitivity and specificity for the BVC used to predict violence in the short-term (cut-off ≥2).

Figure 4

Forest plots of pooled sensitivity and specificity for the BVC used to predict violence in the short-term (cut-off ≥3).

4.4.5. Health economic evidence

No studies assessing the cost effectiveness of prediction instruments for violent and aggressive behaviour by mental health service users in health and community care settings were identified by the systematic search of the economic literature. Details on the methods used for the systematic review of the economic literature are described in Chapter 3.

A case identification model that would model the health and cost consequences of risk prediction of violent and aggressive incidents by mental health service users was considered to be useful; nevertheless, the available clinical and cost data were not of sufficient quality to populate an informative model.

Economic evidence statement

No relevant economic evaluations were identified. Moreover, it was not possible to undertake economic modelling in this area.

4.5. LINKING EVIDENCE TO RECOMMENDATIONS

4.5.1. Risk factors and prediction of violence and aggression

Relative value placed on the outcomes considered

For the review of risk factors, the association between a risk factor and the occurrence of violence/aggression (controlling for other factors) was the outcome of interest. Therefore, only studies that used a multivariate model to determine factors that were independently associated with violence were included. For the review of prediction instruments, sensitivity and specificity of each instrument was primarily used to assess test accuracy. In addition, the AUC and negative and positive likelihood ratios were examined.

Trade-off between clinical benefits and harms

For the review of risk factors, 7 studies (out of 13) with a total of just under 4000 participants were included in the analysis. Of these, 5 included adult participants in an inpatient setting and 2 included adult participants in a community setting. All but 1 study, which was conducted in Taiwan, were conducted in Westernised countries. Most participants were diagnosed with schizophrenia or bipolar disorder and, on average, two-thirds were male.

In inpatient settings for adults, the most notable finding was the paucity of evidence from studies that used multivariate models to establish which factors were independently associated with violence and aggression. With regard to demographic and premorbid factors only age and gender were included in more than 1 study, and no conclusion could be reached based on the evidence. Regarding criminal history factors, no individual factors were included in more than 1 study. Nevertheless, the evidence did support previous reviews, suggesting that recent and lifetime history of violence is an independent risk factor. With regard to psychopathological risk factors, again, few factors were included in more than 1 study, but diagnosis of schizophrenia and later onset of a psychotic disorder were associated with increased risk. With regard to treatment-related factors, 2 studies suggested that the duration of hospitalisation was unlikely to be a risk factor, and the largest study reported referral by a crisis intervention team, referral by home staff (for those living in supported housing) and involuntary admission were independent risk factors. In community settings for adults, the only factors demonstrated to be risk factors in both studies were history of being victimised and recent drug use. Other risk factors demonstrated in 1 study were history of violence – for women only – and conviction for a non-violent offence. In women, African–Caribbean ethnicity was also an independent risk factor for violence. Based on this evidence and the GDG's expert opinion, several recommendations were made about assessing and managing the risk of violence and aggression (see discussion below under ‘Other considerations’ for further rationale).

For the review of prediction instruments, the evidence suggested that the BVC using a cut-off of 2 or more has the best trade-off between sensitivity and specificity.

Pooled likelihood ratios indicate that the test is relatively accurate. The BVC combined with a visual analogue scale (cut-off ≥7) has similar sensitivity and specificity. The DASA has poorer accuracy than the BVC, but still has good sensitivity and moderate specificity. The HCR-20 Clinical Scale has good sensitivity but only low specificity. These findings need to be contrasted with unstructured clinical judgement, which was shown to have poor sensitivity even when both a doctor and nurse agreed about each service user's risk of short-term violence. The GDG agreed that prediction instruments should not be used to grade risk (for example, as low, medium or high), but rather as part of an approach to monitor and reduce incidents of violence and aggression and to help develop a risk management plan in inpatient settings. Recommendations were then drafted in light of the knowledge that incorrectly assessing a service user as high risk could harm the therapeutic relationship.

Trade-off between net health benefits and resource use

Because the costs and consequences of violent events are substantial, there are clear resource and quality of life implications associated with prediction instruments that allow prevention and containment.

From the clinical review, the use of prediction instruments based on risk factors does appear to offer utility over clinical opinion alone. Given the potentially serious clinical and cost consequences of violent and aggressive incidents, any improvement in the management of an event due to prescience is considered likely to be cost effective.

Quality of the evidence

For the review of risk factors, across the inpatient studies and across the community studies, the samples do appear to represent the population of interest and therefore the risk of bias associated with this factor was judged to be low. However, all but 1 inpatient and 1 community study were conducted outside the UK. With regard to loss to follow-up, poor reporting made it difficult to judge whether any loss was unrelated to key characteristics of the sample. With regard to measurement of risk factors and violence and aggression, the potential for bias was judged to be low because of the methods used. With regard to confounders and statistical analysis, only studies using an appropriate multivariate analysis were included in the evidence, and therefore the risk of bias was judged to be low.

For the review of prediction instruments, for all studies included in the statistical analysis the risk of bias was generally low. However, in all studies the reference standard was assessed by staff who also completed the instrument being investigated. This issue is well discussed in the literature and potentially leads to a false positive test rate that is exaggerated because the observed behaviour itself will usually lead to staff taking action to prevent violent behaviour.

Other considerations

Taking into account the evidence presented in this chapter, the GDG also reviewed the recommendations from the previous guideline and judged, based on their expert opinion, that several recommendations were still relevant and of value but would need redrafting in the light of the current context, a widening of the scope and the latest NICE style for recommendations.

Following this approach, the GDG agreed, using consensus methods described in Chapter 3, a framework for anticipating violence and aggression in inpatient wards. It was also agreed that it is good practice to undertake risk assessment and risk management using a multidisciplinary approach, and that the staff who undertake assessments of the risk of violence and aggression should be culturally aware. The GDG also saw the benefit of recommending that risk assessments and management plans should be regularly reviewed in the event that the nature of the risk had changed. Finally, following discussion about modifications to recommendations about risk assessment for community and primary care settings, the GDG wished to emphasise that staff working in these settings should share information from risk assessment with other services, partner agencies such as the police and probation services, and with the person's carer if there are risks to them.

Following the stakeholder consultation, the GDG added a recommendation for staff to consider offering psychological help to develop greater self-control and techniques for self-soothing. A similar recommendation had been developed for children and young people and a stakeholder requested that this recommendation be included for adults.

4.6. RECOMMENDATIONS

Anticipating and reducing the risk of violence and aggression

A framework for anticipating and reducing violence and aggression on inpatient psychiatric wards

Use the following framework to anticipate violence and aggression in inpatient psychiatric wards, exploring each domain to identify ways to reduce violence and aggression and the use of restrictive interventions.

• Ensure that the staff work as a therapeutic team by using a positive and encouraging approach, maintaining staff emotional regulation and self-management (see recommendation 5.7.1.36) and encouraging good leadership).

• Ensure that service users are offered appropriate psychological therapies, physical activities, leisure pursuits such as film clubs and reading or writing groups, and support for communication difficulties.

• Recognise possible teasing, bullying, unwanted physical or sexual contact, or miscommunication between service users.

• Recognise how each service user's mental health problem might affect their behaviour (for example, their diagnosis, severity of illness, current symptoms and past history of violence or aggression).

• Anticipate the impact of the regulatory process on each service user, for example, being formally detained, having leave refused, having a failed detention appeal or being in a very restricted environment such as a low-, medium- or high-secure hospital.

• Improve or optimise the physical environment (for example, use unlocked doors whenever possible, enhance the décor, simplify the ward layout and ensure easy access to outside spaces and privacy).

• Anticipate that restricting a service user's liberty and freedom of movement (for example, not allowing service users to leave the building) can be a trigger for violence and aggression.

• Anticipate and manage any personal factors occurring outside the hospital (for example, family disputes or financial difficulties) that may affect a service user's behaviour.

Assessing and managing the risk of violence and aggression

When assessing and managing the risk of violence and aggression use a multidisciplinary approach that reflects the care setting.

Before assessing the risk of violence or aggression:

• Take into account previous violent or aggressive episodes because these are associated with an increased risk of future violence and aggression.

• Do not make negative assumptions based on culture, religion or ethnicity.

• Recognise that unfamiliar cultural practices and customs could be misinterpreted as being aggressive.

• Ensure that the risk assessment will be objective and take into account the degree to which the perceived risk can be verified.

Carry out the risk assessment with the service user and, if they agree, their carer. If this finds that the service user could become violent or aggressive, set out approaches that address:

• service-user related domains in the framework (see recommendation 4.6.1.1)

• contexts in which violence and aggression tend to occur

• usual manifestations and factors likely to be associated with the development of violence and aggression

• primary prevention strategies that focus on improving quality of life and meeting the service user's needs

• symptoms or feelings that may lead to violence and aggression, such as anxiety, agitation, disappointment, jealousy and anger, and secondary prevention strategies focusing on these symptoms or feelings

• de-escalation techniques that have worked effectively in the past

• restrictive interventions that have worked effectively in the past, when they are most likely to be necessary and how potential harm or discomfort can be minimised.

Consider using an actuarial prediction instrument such as the BVC (Brøset Violence Checklist) or the DASA-IV (Dynamic Appraisal of Situational Aggression – Inpatient Version), rather than unstructured clinical judgement alone, to monitor and reduce incidents of violence and aggression and to help develop a risk management plan in inpatient psychiatric settings.

Consider offering service users with a history of violence or aggression psychological help to develop greater self-control and techniques for self-soothing.

Regularly review risk assessments and risk management plans, addressing the service user and environmental domains listed in recommendation 4.6.1.1 and following recommendations 4.6.1.3 and 4.6.1.4. The regularity of the review should depend on the assessment of the level of risk. Base the care plan on accurate and thorough risk assessments.

If service users are transferring to another agency or care setting, or being discharged, share the content of the risk assessment with staff in the relevant agencies or care settings, and with carers.

Managing violence and aggression

After a risk assessment has been carried out, staff working in community and primary care settings should:

• share the risk assessment with other health and social care services and partner agencies (including the police and probation service) who may be involved in the person's care and treatment, and with carers if there are risks to them

• be aware of professional responsibilities in relation to limits of confidentiality and the need to share information about risks.

4.7. RESEARCH RECOMMENDATIONS

What is the effect of detention under the Mental Health Act on rates of incidence of violence and aggression in inpatient psychiatric wards?

Are Safewards and/or short term risk assessment effective ways to reduce rates of inpatient aggression?

Examples include Christopher Clunis, a service user with schizophrenia, who killed Jonathan Zito in London in 1992. The subsequent inquiry (Ritchie et al., 1994) identified multiple failures in the care provided to Clunis, including poor communication, lack of continuity and reluctance to provide services to him. Another example is Michael Stone, an individual with psychopathic disorder who killed Lin Russell and her 6-year-old daughter Megan in Kent in 1996 while her 9-year-old daughter Josie survived with severe head injuries. This incident significantly contributed to the introduction of services for people with ‘dangerous and severe personality disorders’ (Völlm & Konappa, 2012).

Here and elsewhere in the guideline, each study considered for review is referred to by a study ID (primary author and date of study publication, except where a study is in press or only submitted for publication, then a date is not used).

Here and elsewhere in the guideline, each study considered for review is referred to by a study ID (primary author and date of study publication, except where a study is in press or only submitted for publication, then a date is not used).

When assessing a client what is the most important predictor of potential for violence?

What is a characteristic of unit culture that predicts client violence?

Which personality trait is associated with aggressive behavior?

Which personality trait is associated with aggressive behavior quizlet?