Inpatient Violence

The threat of violence is a major concern for all individuals working or receiving treatment in an inpatient psychiatric setting [1]. For example, Silver and Yudofsky [2] recorded over 3000 episodes of aggressive behavior (both verbal and physical) in a 20-month period in just 30 psychiatric inpatients. Fellow patients are not the only victims of such assaults. According to the Bureau of Labor Statistics, in 1999, 2637 nonfatal assaults on hospital workers occurred, which translates to a rate of 8.3 assaults per 10,000 workers. This contrasts to a rate of assault of 2 per 10,000 workers in all other industries. The victims of such assaults were typically nursing staff, primarily because they spend the most time with patients. While these numbers include all healthcare employees, the Department of Justice National Crime Victimization Survey for 1993 through 1999 [3] reported that the average annual rate for nonfatal violence for mental health professionals was 68.2 per 1000 compared to a rate of 12.6 for all occupations. While the Occupational Safety and Health Administration (OSHA) provided compelling reasons for these numbers (e.g., increasing use of hospitals by police officers for the care of acutely disturbed violent individuals), the consequences are clear: many mental health professionals will at some point in their careers be the victim of an assault. The resultant loss of work and psychological trauma to both the providers and the patients is significant [4,5]. Developing methods for reducing violence on inpatient psychiatric units is of paramount importance.

The Development of Risk Assessments

One major focus in forensic psychology and psychiatry over the past several decades has been the development of risk assessments to aid in the identification of those individuals most at risk of exhibiting violent behavior. Initially, these instruments were developed primarily to make informed decisions regarding when (or if) offenders could be safely released into the community. So-called “first-generation” risk assessment consisted of unstructured, often uninformed, estimates of dangerousness, which were notoriously fallible and unreliable. This is primarily because mental health professionals placed an increased importance on idiosyncratic factors that may not bear any relationship to violence [6]. Monahan’s monograph [7] documents that when using this approach, mental health professionals were inaccurate 67% of the time. “Second-generation” risk instruments were developed in the 1980s and 1990s and employed statistical analyses to assess the mathematical relationship between person variables (e.g., age, elementary school performance) and the measured outcome (violence or violent offending). Many of these instruments provided probability values that allowed mental health professionals to quantify the risk of recidivism or aggression. Aside from providing an artificial sense of accuracy (some authors suggest that these probability values, when applied to the individual, vary widely [8], although others argue that the mathematical equations to support such arguments are flawed [9]), most included primarily static or historical factors, which are not amenable to change [10,11]. One static factor that has been researched extensively is the construct of psychopathy, typically operationalized by the Hare Psychopathy Checklist–Revised (PCL-R) [12] or one of its derivatives. This construct, when included in the risk instrument, often carries the most weight. However, because many believe the primary purpose of risk assessment is risk management [13,14], static factors in theory yield little information regarding management of risk. Additionally, these second-generation risk assessments were developed to identify those individuals who are more likely to become violent in the long-term. For example, the Violence Risk Appraisal Guide (VRAG) [10,15,16] estimated the probability of violent offending at 7- and 10-year follow-up points. For inpatient psychiatry, static predictors of risk may only be useful if containment is a feasible method of risk management. The use of seclusion and restraints is often viewed as containment and nontherapeutic [17], and their use is increasingly limited in most psychiatric facilities [18]. Because psychiatric hospitals typically endeavor to focus on treatment, not containment, risk instruments with exclusively static predictors were believed to hold little utility for the management of aggressive patients. Long-term risk, which was often the focus of earlier risk instruments, is generally of importance when release is being considered. Short-term risk of aggression is much more salient to administrators of acute psychiatric units and long-term care facilities. Recent research has suggested that factors related to community violence (and therefore representative of long-term risk) may be substantially different than factors related to institutional aggression (i.e., short-term risk), although differences also may be related to the operational definition of violence [19,20].

“Third-generation” risk assessments were developed based on empirical literature describing the relationships between person variables and outcome. These tools contain both static and dynamic factors that place an individual at risk. These guided professional assessments, often termed structured professional judgment (SPJ) tools, contain items that have been empirically linked to aggression or offending (rather than statistically linked) and incorporate dynamic risk factors for two specific reasons: to provide information on how to manage risk and to provide a method for measuring change in risk as a result of treatment. As Hart [14] cogently noted, the goal of risk assessment ultimately is prevention of violence. Thus, assessment methods must include factors that are amenable to intervention [21]. Many SPJ tools include factors more traditionally associated with mental illness (e.g., diagnosis, current symptoms), which are considered more dynamic, rather than antisocial traits, which could be viewed as more historical or static. Douglas and Skeem [21] proposed various dynamic risk factors for further investigation, including impulsivity, anger, and substance use. More recently “fourth-generation” risk tools have been proposed as superior, as these assessments contain items that are dynamic as well as potentially protective, and they place a focus on the linkage between identified risk factors and proposed interventions. However, the utility of these instruments is still unclear [22]. See Table 4.1 for a summary of the development of violence risk assessments.

The development of second- and third-generation risk instruments appears to have produced the desired results: many of these tools have been shown to provide improved accuracy over unguided clinical judgment when used in an appropriate fashion, especially as they relate to community violence. For reviews of the efficacy of risk assessment and its relationship to community violence, see for example Edens and Otto [23], Quinsey et al. [15], or Walters [24]. The strength of the relationship of these risk instruments in general, and psychopathy specifically, to community violence led investigators to extend the examination to the institutional setting. Institutional aggression is a common problem in both general and forensic psychiatric facilities, as well as in correctional facilities. While initial investigations supported a strong positive relationship between the PCL-R (a static risk factor) and institutional aggression [25,26], two meta-analyses have indicated that the PCL-R is not as robust a predictor of institutional aggression as initially believed [20,27]. In addition to concerns regarding the utility of psychopathy within institutions, concerns also have been raised regarding the predictive validity of instruments such as the VRAG and the Historical–Clinical–Risk Management (HCR-20) [28] in relation to institutional violence [19]. Compared to the community violence literature, relatively few studies have examined the efficacy of these measures in identifying violence-prone individuals in inpatient settings. McNiel et al. [29] evaluated the utility of various structured risk instruments in identifying individuals who would exhibit aggression in the short-term (mean length of stay of the sample was 9.5 days). They found that the clinical scale of the HCR-20 was most associated with inpatient violence. In a sample of 97 male offenders, Kroner and Mills [30] found that the VRAG evidenced the strongest relationship with major institutional misconduct, whereas the HCR-20 total score evidenced the lowest, although no differences were statistically significant. In contrast, Doyle et al. [31] found in a study of 87 mentally disordered offenders in a medium security facility that the PCL-SV was superior to the VRAG and the H scale of the HCR-20 in predicting serious aggression. In a sample of 154 forensic inpatients, McDermott et al. [32] found that the clinical and risk management subscales of the HCR-20, as well as hostility as measured by the Brief Psychiatric Rating Scale (BPRS), were most strongly associated with patient-directed aggression, whereas two subscales from the Novaco anger scale were most strongly associated with staff-directed aggression. Vitacco et al. [33] found that both static and dynamic factors predicted inpatient aggression, although none exhibited incremental validity beyond any other. In a meta-analysis of 95 studies examining the utility of the PCL in both community and institutional settings, Leistico et al. [34] found that the PCL exhibited greater utility in a psychiatric inpatient sample, as compared to an incarcerated correctional sample. The results of these studies suggest that there is some utility to including both static and dynamic risk factors in risk instruments as they relate to institutional aggression.

Consistent with Douglas and Skeem’s suggestion [21] and McDermott et al.’s research [32], the extant literature offers considerable support for the contention that anger and hostility are risk factors for interpersonal aggression. For example, research has shown positive associations between anger scales and institutional misconduct among violent male offenders in prison [35,36]. In the MacArthur study of mental disorder and violence [37], anger was a potent predictor of community violence, so much so that it was incorporated into the Classification of Violence Risk (COVR) [38] as one of the 44 risk factors. More recently, Ullrich et al. [39] found that anger was a critical component in the pathway between delusions and violence. Impulsivity also has been associated with a greater likelihood of acting out in controlled environments [40–42]. Recently, a prospective examination of the relationship of second- and third-generation risk instruments with institutional violence [32] found that one type of patient was responsible for the majority of the aggression exhibited toward both staff and other patients in a forensic facility: patients who scored high on risk instruments (including the PCL-R, the VRAG, and the HCR-20) and who also scored high on self-report measures of anger and impulsivity. Patients who scored high only on the risk instruments but not on measures of anger and impulsivity exhibited significantly lower rates of aggression. These data provide further support that both dynamic and static risk factors are useful in identifying potentially aggressive patients.

A Typology of Aggression

In addition to categorizing aggression as staff- or patient-directed, there is also evidence that institutional aggression can be categorized according to precipitating events and that the factors associated with each type of aggression may vary. In a study conducted in a state psychiatric hospital in New York, three primary motivations for assault were described: (1) impulsive – an assault committed in response to an immediate provocation and associated with agitation and loss of emotional control; (2) planned – a controlled assault committed for a specific goal; and (3) psychotic – assaults committed as a consequence of delusions, hallucinations, and/or disordered thinking [43]. Symptoms of mental illness, especially a major mental disorder, are necessarily associated with psychotic aggression by definition; psychotic symptoms are less likely to be linked to planned/predatory aggression. Quanbeck et al. [44] examined almost 1000 acts of aggression exhibited by both civil and forensic patients and found that a similar trichotomy applied, which they termed impulsive, organized, and psychotic. They also were able to delineate the underlying motivation for each type of aggression. For example, an act of organized aggression may be exhibited to achieve social dominance, whereas an act of impulsive aggression may be precipitated by the patient being denied a desirable item. Using this categorization scheme, impulsive/reactive assaults comprised the largest number of observed incidents of aggression (46%) [44]. See Table 4.2 for a description of the typology of violence. Studies have been conducted to determine if traditional second- and third-generation risk instruments hold more utility in identifying individuals who are more likely to exhibit certain types of aggression. Some research suggests that planned/predatory aggression is more strongly related to psychopathic and antisocial attitudes [45,46]. In a sample of 152 male forensic patients, Vitacco et al. [47] found that anger and symptoms of mental illness, but not psychopathy, were associated with reactive aggression. In contrast, psychopathy, hostility, and anger were most associated with instrumental aggression. McDermott et al. [48] found similar results: in a sample of 238 predominantly male forensic patients, psychopathy and anger were most strongly related to predatory aggression; the clinical and risk management scales of the HCR-20 and anger were most associated with impulsive aggression. Not surprisingly, symptoms of mental illness and impulsivity were most associated with psychotically motivated aggression. Interestingly, these relationships only held true when follow-up was limited to six months; longer follow-up attenuated most of the relationships.

In psychiatric facilities, the temporal occurrence of aggression is extremely important and may, at least in part, be associated with the type of aggression exhibited. The study by McDermott et al. [48] demonstrated that various instruments may be differentially associated with various types of aggression at varying time points in the hospitalization. For long-term care facilities, imminent (defined as within the next 24 hours), short-term (variously defined as between one and six months), and long-term (defined in years) are all relevant in the treatment and management of psychiatric patients. For acute care inpatient psychiatric units, imminent aggression is most relevant, as most inpatient stays are extremely brief. Thus, the predictors of imminent aggression may be different from short-term aggression, which may also be quite different from the predictors of longer-term recidivism and offending. As an example, Woods and Almvik [49] found that the Brøset Violence Checklist (BVC) [50], which contains six items related to patient behavior/affect, was effective in identifying individuals at higher risk of aggression in the 24-hour period following the assessment. Other authors have determined that predictors of short-term versus long-term aggression may differ widely, and are especially dependent on the type of aggression [48].

Although most standard risk instruments have been developed on forensic samples, recent research suggests that many risk instruments are applicable in both civil and forensic settings and that perhaps more important is the type of patient and the time frame under which the assessments are employed. This stands to reason, as most forensic commitments require a severe mental illness that has led to the impairment in legal reasoning. There are several validated risk assessment tools that were developed specifically to assess violence risk in the acute inpatient setting. Examples are the McNiel–Binder Violence Screening Checklist (VSC) [50], the BVC [51], and the Dynamic Appraisal of Situational Aggression (DASA) [52].

The VSC, which was developed at a university-based, locked, short-term psychiatric unit, is a screening tool that uses four variables to predict aggression and violence within 72 hours of inpatient admission. These four variables are a history of physical attacks or fear-inducing behavior within two weeks prior to admission, absence of suicidal behavior within two weeks of admission, schizophrenic or manic diagnosis, and male gender. An individual having a positive score on the VSC is 1.97 times more likely to engage in a physical attack or fear-inducing behavior. In addition, the VSC has demonstrated an ability to correctly classify 65% of patients as violent or nonviolent in a population of patients with a base rate of violent behavior of 41%.

The BVC was developed at the Norwegian maximum security unit at Brøset, which houses patients with mental disorders who were felt to be highly dangerous and difficult to manage. The checklist consists of six behaviors found to be most common during a 24-hour period prior to a violent incident, namely confusion, irritability, boisterousness, physical threats, verbal threats, and attacks on objects. The BVC has demonstrated 63% accuracy in predicting that violence will occur within the next 24 hours and 92% accuracy in predicting that violence will not occur. A randomized, controlled trial based in a Swiss psychiatric facility utilized the BVC as part of a violence risk reduction intervention during acute admission that resulted in a 41% risk reduction in severe events of inpatient aggression and a 27% risk reduction in the need for coercive measures against patients.

The DASA is a seven-item instrument, which contains two items from the HCR-20, two from the BVC, and three based on the authors’ own research. In the development study [52], the DASA evidenced an area under the curve (AUC) of 0.82 with all seven items combined. The odds ratio of a patient scoring 7 (the highest score) compared to one scoring a 0 of committing an act of aggression within 24 hours was 29 (the patient was 29 times more likely to commit an act of aggression).

Many of these instruments that are used to identify patients likely to exhibit imminent aggression rely on observation of escalating behaviors. This suggests that these tools provide the greatest utility in identifying individuals at risk of exhibiting psychotically motivated or impulsive aggression. The SPJ instruments may be of more utility in identifying those individuals who are at risk of aggression later in their course of treatment, and again, aggression that may be viewed as impulsive or psychotically motivated. Static measures of risk appear to hold the greatest utility for identifying those individuals who are most likely to exhibit predatory aggression. In a direct assessment of this, Grann et al. [53] found that static risk instruments were most efficacious in individuals with personality disorders. These same risk instruments held little utility for individuals with major mental disorders. They suggested that dynamic factors may be more useful for such individuals.