Treatment Nonadherence

Medication nonadherence predicts violence among psychiatric patients (Bartels et al. 1991; Monahan et al. 2001). Patients’ positive perceptions of their treatment need and treatment effectiveness are associated with reduced odds of violence (Elbogen et al. 2006). Both medication adherence and treatment participation fluctuate over time (Svedberg et al. 2001) and may have both direct and indirect relationships with violence.

Approaches to Violence Risk Assessment

As the foundational knowledge underlying violence risk assessment has evolved over the past few decades, so have the methods to integrate and apply this knowledge in individual cases.

Unstructured Clinical Judgment

Historically, the primary professional approach to violence risk assessment has been unstructured clinical judgment. As Monahan (2008) explained, “In unstructured assessment, risk factors are selected and measured based on the mental health professional’s theoretical orientation and prior clinical experience. What these risk factors are, or how they are measured, might vary from case to case depending on which seem most relevant to the professional doing the assessment. At the conclusion of the assessment, risk factors are combined in an intuitive or holistic manner to generate an overall professional opinion about a given individual’s level of violence risk” (p. 19).

The process in unstructured clinical judgment is not necessarily transparent, objective, accurate, or reliable (Lidz et al. 1993; Monahan 1981, 2008). Absent structure, assessments are vulnerable to inconsistency across clinicians and even by the same clinician across different cases (Guy et al. 2015). Unstructured approaches have performed particularly poorly when contrasted with the highly structured actuarial approach.

Actuarial Assessment

Actuarial approaches (also known asmechanical, statistical, or mathematical approaches) use a research-derived formula to estimate the probability of future violence. Actuarial approaches define which data should

be considered and what algorithm should be used to weigh and combine that data, leading to a fixed conclusion about risk (Grove and Meehl 1996; Meehl 1954). The most familiar examples of actuarial approaches, for most people, are the algorithms that insurance companies use to set insurance rates.

Researchers have repeatedly compared the unstructured clinical judgment approach with actuarial approaches and have concluded that actuarial approaches are far superior (Dawes et al. 1989; Grove and Meehl 1996; Grove et al. 2000; Hanson and Morton-Bourgon 2009). For example, the authors of a comprehensive meta-analysis comparing clinical versus actuarial prediction across many disciplines concluded that “one area in which the statistical method is most clearly superior to the clinical approach is the prediction of violence” (Ægisdóttir et al. 2006, p. 368).

Most actuarial measures are easy-to-use tools, developed from a specific data set, that allow users to code—for a particular individual—certain clearly defined risk factors that were measured in the original data set, and then examine the frequency of a particular outcome (e.g., violence) among individuals in the data set with the same risk factors, or the same number of identified risk factors. Afterward, a clinician offers a structured conclusion such as, “Mr. Smith has X risk factors, making him similar to group Y, of whom 27% went on to commit violence.” The perceived objectivity of actuarial approaches and supportive research have been so compelling that some scholars argue that sole reliance on actuarial measures is the only appropriate means to assess risk (Quinsey et al. 2006).

Most clinicians in routine practice appear reluctant to adopt actuarial violence risk assessment instruments (Elbogen et al. 2002; Hilton et al. 2006; Monahan 2008) because they do not allow consideration of some case- specific data, particularly unique or dynamic variables. Also, actuarial instruments have limited utility for guiding specific interventions or risk management strategies.

Structured Professional Judgment

Other modern violence risk assessment measures reflect an approach labeled structured professional judgment (SPJ) (Webster et al. 1997). SPJ is intended to capture the strengths but minimize the weaknesses of both actuarial and clinical judgment approaches. Tools based on the SPJ model

delineate research-identified risk factors (just as actuarial measures do) but rely on the clinician’s judgment to gather, weigh, and combine these risk factors into a final risk formulation. Like actuarial measures, they tend to be m o r e reliable, transparent, and accurate than unstructured assessment; however, unlike actuarial approaches, they allow clinicians the flexibility to consider factors beyond the instrument and/or weigh some factors as more important than others, inviting clinical judgment.

SPJ instruments comprise 1) a list of static and dynamic risk factors to consider and 2) a scheme for coding these factors (usually as absent, partially present, or present). SPJ instrument manuals usually provide recommendations for collecting information, determining final opinions, and communicating these opinions about violence risk. Final opinions about risk are not determined solely by summing the risk factors; clinicians may conclude for qualitative reasons that actual risk is higher or lower than a simple tally of risk factors would suggest (Douglas et al. 2013; Webster et al. 1997). Final risk opinions are not communicated numerically but are conceptualized and communicated in a categorical manner (e.g., low, moderate, or high risk) to reflect the clinician’s degree of concern about future violence.

Using Instruments to Assess Violence Risk Assessments

Historically, psychiatrists have not used risk instruments often, perhaps because few instruments were conducive to fast-paced, routine psychiatric practice. However, even psychiatric authorities increasingly emphasize that structured assessment approaches outperform unstructured approaches (Buchanan et al. 2012). Structured approaches are particularly important in forensic contexts, in which assessments must be transparent and defensible. Indeed, most modern forensic or correctional violence risk assessments are facilitated by a structured instrument (Singh et al. 2014).

Instruments can be categorized by the degree to which they structure four key components, or steps, of the risk assessment process: 1) identifying risk factors, 2) measuring or scoring the risk factors, 3) combining risk factors, and 4) producing a final risk estimate (Monahan 2008; Skeem and Monahan

2011). A written list of research-identified risk factors adds structure to the first step but not to the subsequent steps. Popular SPJ approaches such as the Historical Clinical Risk Management–20 (HCR-20; Webster et al. 1997) structure two of these four components (i.e., identifying risk factors and scoring the risk factors). At the far end of the continuum, the highly structured, purely actuarial measures, exemplified by the Violence Risk Appraisal Guide (VRAG;Quinsey et al. 1998, 2006), structure all four components of the process.

Historical Clinical Risk Management–20

The HCR-20 (Webster et al. 1997), now in its third version (Douglas et al. 2013), is the most widely used and researched violence risk assessment approach using the SPJ model and is probably the most widely used violence risk assessment instrument of any sort (Singh et al. 2014). The HCR-20 was developed to assess risk of violence among adult male and female civil psychiatric patients, forensic psychiatric patients, and criminal offenders (Douglas et al. 2013).

The HCR-20 comprises 20 risk factors grouped into three domains: Historical (past), Clinical (present), and Risk Management (future). Extensive literature supports the instrument’s reliability (across raters) and predictive validity (with respect to predicting violent outcomes). Clinician risk judgments based on the HCR-20 appear to perform as well as or better than those based on other risk instruments, according to several large-scale meta-analytic studies (Campbell et al. 2009; Fazel et al. 2012; Guy 2008; Singh et al. 2011).

Classification of Violence Risk

The Classification of Violence Risk (COVR; Monahan et al. 2005, 2006) was developed with data from the seminal MacArthur Violence Risk Assessment Study (Monahan et al. 2001). The COVR is an interactive computer software program that guides the clinician through a chart review and brief patient interview necessary to measure 40 risk factors and calculate a risk estimate following an “iterative classification tree” methodology (Monahan 2010; Monahan et al. 2005). Finally, the COVR generates a report that places the examinee’s violence risk into one of five categories based on the violence

rates among subsamples of patients in the MacArthur study (e.g., a 1% rate of violence in the lowest-risk subsample, and a 76% rate of violence in the highest-risk subsample). Nevertheless, the clinician using the instrument, not the instrument itself, is responsible for the final risk estimate. The clinician should begin with the instrument-generated violence risk category but also consider the possibility of any factors (beyond those in the instrument) that may raise or lower risk before offering a final risk estimate and developing a risk management plan.

Violence Risk Appraisal Guide

The VRAG (Quinsey et al. 1998, 2006) is a 12-item actuarial instrument developed through an extensive program of research with mentally ill offenders in Canada. Researchers coded dozens of potential risk factors from the institutional files of a maximum-security forensic psychiatric hospital, and then followed patients for an average of 7 years after release, documenting new criminal charges for violence (or returns to the hospital for similarly violent behavior). The instrument requires evaluators to code 12 risk factors, which are then statistically weighted and summed to produce an overall estimate of violence risk. This instrument-produced estimate must be the only final risk estimate; the instrument authors warn that “clinical judgment is too poor to risk contaminating” an objective actuarial estimate (Quinsey et al. 2006, p. 197).

Instrument Selection

No credible argument or data demonstrate that completely unstructured, unguided clinical judgment is better than more structured approaches. However, much room remains for reasonable disagreement about which structured approach (i.e., which instrument) is most accurate, and much research has attempted to solve this conundrum (Heilbrun et al. 2010; Singh et al. 2011). A review of such literature is beyond the scope of this chapter, but it can be fairly summarized by stating that no single structured violence risk assessment tool consistently outperforms all others (Campbell et al. 2009; Fazel et al. 2012; Singh et al. 2011).

Performing the Violence Risk Assessment

Performing the Violence Risk Assessment

As in all forensic evaluations, the evaluator should begin the violence risk assessment with a clear referral question and authority to proceed (e.g., a court order or engagement letter from counsel). Then the clinician should gather all available, relevant collateral data before interviews with the examinee.

The clinician’s goals usually involve violence prevention more than violence prediction. Therefore, the aim of the assessment is to gain a rich understanding of the examinee’s violence risk sufficient to plan risk management strategies. As the examiner proceeds through the evaluation, the distinction between risk status and risk state provides a helpful framework for organizing risk-relevant information (Douglas and Skeem 2005); the former is crucial to understanding an individual’s risk relative to others, and the latter is crucial to understanding changes in an individual’s risk over time.

Risk Status

Generally, risk status involves a patient’s risk of violent behavior relative to other individuals in a particular population or context (Douglas and Skeem 2005). Some examinees will always remain at higher risk status and warrant closer risk monitoring because of unalterable historical characteristics, such as past violence.

Even at the earliest stages of evaluation, a clinician can obtain information about a patient’s psychiatric and criminal history, as well as other personal data, which can be used to identify the patient’s population or comparison group for the purpose of considering base rates of violence. This referral information may include some basic static risk factors for violence. During interviews, the clinician can elicit more information about a patient’s risk status. By inquiring about the details of psychiatric symptoms, substance abuse, and especially violence history, the clinician further assesses risk status. The clinician must ask questions that are specific to past violence. Such questions include a thorough review of all past instances of violence:

What were the nature, type, frequency, and severity? Who were past victims?

What was the context or setting for the violence?

What events preceded and followed the violence?

How recent was the last instance of violence, and is there any evidence of escalation?

Eliciting all information about past violence is essential to understanding the contexts and situations in which the examinee could most likely commit violence in the future. These details not only help inform risk status but also plant the seeds of future risk management strategies. Finally, the clinician should ask about instances in which the examinee was nearly violent but did not proceed with violence. The answers may provide clues to the examinee’s strengths and potential risk management strategies that a clinician can use later.

When forming an initial estimate of risk status, the clinician should (in most circumstances) use one of the many well-researched instruments for assessing violence risk, as described in the previous section. An underrecognized benefit of structured risk instruments is helping clinicians to avoid an overemphasis on psychiatric symptoms and to consider the much broader range of (nonpsychiatric) risk factors for violence.

Risk State

In contrast to risk status, risk state refers to a person’s current violence risk compared with his or her own risk in the past. In other words, risk state involves the “individual’s propensity to become involved in violence at a given time, based on particular changes in biological, psychological, and social variables in his or her life” (Douglas and Skeem 2005, p. 349). Therefore, assessing risk state involves a focus on the examinee’s current clinical status, for example:

Are there changes in the psychiatric symptoms that seem most relevant to violence risk?

Is substance use increasing?

Has conflict with family escalated?

In many ways, these considerations are common in routine psychiatric practice for assessing improvement or decline in clinical functioning and intervening appropriately. Forensic evaluators, however, consider these

clinical changes as they relate to violence potential and then explicitly explore with examinees the prospect of violence.

Optimal assessment involves not only addressing an examinee’s current risk state but also anticipating factors that would change that risk state. For example, the examiner should consider questions such as these: Is the examinee’s sobriety tenuous? Is he or she involved in a volatile relationship that could escalate toward violence? Is an otherwise high-risk individual in a stable and protective relationship such that if the person lost the relationship, he or she would likely resume violence without the stability the relationship provides?

Evaluators may perceive a change in risk state because the examinee conveys—whether to the evaluator or to others—a threat of violence or a desire to harm someone. Patients may convey threats in a manner that is overt and intentional, or inadvertent and accidental. In either scenario, authorities use the term “threat leakage” to describe situations in which an individual conveys to a third party the intent to harm a target (Meloy and O’Toole 2011). In addition to threatening statements or articulated desires, threat leakage may include behaviors that leave a clinician concerned that the patient poses a threat to one or more particular victims. For instance, an examinee who loses a relationship or a job, increasingly ruminates on a grievance against the ex-partner or ex-supervisor who caused the loss, and increasingly knows the whereabouts of that person, may pose a threat even if he or she has not explicitly articulated a threat.

Conclusion

The field of violence risk assessment has developed extensive data regarding the rates and correlates of violence, particularly among individuals with mental illness. Clinicians now have a wealth of knowledge to inform risk assessments and a range of well-researched assessment techniques to employ. They need not rely on the unstructured and impressionistic judgments that were common in the past. Moving forward, the challenge will be to employ these best practices in consistent and transparent ways that help courts and facilities employ the best risk management practices.