 |
 |
|
|
|
In 1990 President George HW Bush proclaimed the next 10 years as the “decade of the brain” (1). The proclamation’s purpose was “to enhance public awareness of the benefits to be derived from brain research” (1).The year 1990 also marks the beginning of the contemporary era of research on mental illness and violence. However, despite significant advances in knowledge about risk assessment and the correlates of violence, a 1996 study of public opinion demonstrated increasing public consciousness of an association between mental illness and “dangerous[ness] to self or others” (2). The paradigm of dangerousness evolved in the 1960s and 1970s from the confluence of at least 3 sociopolitical movements related to psychiatry and from the legal cases that highlighted them (3). The first of these was the deinstitutionalization movement, which, in creating an alternative option to hospital care, emphasized the need for criteria to decide between inpatient and outpatient care. The second movement was the increased use of voluntary hospitalization for psychiatric patients. In 1971 most psychiatric patients in the US were for the first time hospitalized voluntarily (4). With the advent of the popular use of voluntary hospitalization came the demand for involuntary hospitalization criteria that could withstand the increased scrutiny levelled at involuntary commitments. The third variable was the civil rights movement, which spawned an increase in libertarian advocacy for persons with mental illness; the highest possible standards were sought for the involuntary deprivation of liberty occasioned by civil commitment. Various civil commitment cases confirmed repudiation of the parens patriae standard and reliance on government police powers and the dangerousness standard (according to which dangerousness to self or others becomes the principal determinant of eligibility for involunatry commitment) (5–8). Despite widespread adoption of the dangerousness standard in commitment proceedings and extension of that standard to psychiatric outpatients (9–11), capital defendants (12,13), and insanity acquittees (14), no data supported the idea that psychiatrists or other mental health professionals could reliably determine future dangerousness. In fact, Monahan’s 1981 review of the literature demonstrated that mental health professionals were wrong in 2 out of 3 attempts to predict dangerousness (15). Prominent forensic mental health professionals (16–19) and the American Psychiatric Association (20) attempted to critique this new paradigm and modify its evolution in mental health care. These efforts were mostly unsuccessful. One result has been that access to both public and private psychiatric inpatient beds has been skewed toward patients representing a “danger” (18,21). Another has been increasing expectations for psychiatrists to function as agents of social control, as approved by the US Supreme Court in Kansas v Hendricks (22). Since Monahan’s 1981 review, however, our scientific knowledge about violence and mental illness has expanded. We clearly have both a better understanding of epidemiologic risk factors for violence and a better-than-chance ability to predict violence (23). In the early to mid 1990s, the field evolved from discussing the prediction of violence toward the assessment of risk for violence, a clarification that is faithful to scientific advances in the field (24). Nevertheless, these advances have in many ways aided neither clinicians nor patients. This paper attempts to outline the extent and limitations of our current knowledge about the correlates of violence and the accuracy of our risk assessments. It also comments on how clinicians today might process this knowledge and use it in practice. Contemporary Research on Correlates of ViolenceEarly Ideas The major advances in knowledge about mental illness and violence that characterize contemporary research arguably began in 1990, with the publication of data from Swanson and colleagues’ large epidemiologic study (25). Before then, most studies attempted either to correlate rates of mental illness among prisoners or pretrial detainees with rates of mental illness among the general population or to correlate arrest rates among persons with mental illness with arrest rates in the general population. These studies were criticized for being methodologically flawed by labelling phenomena, which artificially elevated the correlation between mental illness and violence (26). The prison studies were criticized for applying medical labels to criminal behaviour (the “medicalization of criminal behaviour”). The arrest-rate studies were criticized for applying criminal labels to symptom behaviour (the “criminalization of mental illness”). Advocates for individuals suffering from mental illness argued that there was no association between mental illness and violence that was not explained by such flaws and that this association was due merely to centuries-old stigmatizing of persons with mental illness. Swanson and others took an epidemiologic approach that avoided these previously cited methodological flaws. They used self-report data from the Epidemiological Catchment Area (ECA) Survey to assess the occurrence of violence among individuals with no mental disorder and among those diagnosable with various disorders. Swanson and others found that major mental disorder correlated with at least a fivefold increase in rates of reported violence over a 1-year period (approximately 10% to 13%), compared with rates among individuals with no disorder (2%), and tenfold and higher increases in violence among those with various substance abuse disorders (approximately 19% to 35%). Regarding sociodemographic variables, the highest rate of violence (16%) occurred in young men aged 18 to 24 years in the lowest socioeconomic group (25). In 1992 Link and others published similar results from a different epidemiologic survey. They found that differences in the rate of violence among patient groups and between patient and community control groups could not be fully explained by sociodemographic and environmental factors (such as homicide rates that varied by census tract). They found that active psychosis explained these differences and that individuals who were identified as patients but who were not currently suffering from psychosis were at no greater risk for violence than the average community control population. They also cautioned that the risk for violence correlated to mental illness in their sample was less than the risk correlated to young age and male sex (26). Psychosis and Violence These findings sparked the search for more specific data about the nature of psychosis and its relation to risk for violent behaviour. Link and Steuve hypothesized that specific types of paranoid psychotic symptoms would make violence more likely as a rational response to an irrationally perceived external threat; they referred to this as the “rationality-within- irrationality” hypothesis. To test this, they reexamined specific psychotic symptoms from their survey data, characterized as “threat/control/override” (TCO). This concept referred, respectively, to the beliefs that others meant to do one harm, that others could control one’s thoughts, or that others could put thoughts into one’s head. The researchers discovered that such symptoms were better predictors of violence than were psychotic symptoms in general (27). Swanson and others replicated this finding in 1996, using their data from the ECA Survey. They found that respondents with TCO symptoms were twice as likely to report violence as those with other psychotic symptoms, 6 times as likely to report violence as those with no mental disorder, and, when combined with substance abuse, 8 times more likely to report violence than those without a mental disorder (28). In 1997 Link and colleagues published a follow-up study examining the relation between threat symptoms and control–override symptoms and the occurrence of reported violence. Using epidemiologic data from Israel, they found that the 2 sets of symptoms independently and significantly predicted violence (29). Toward the end of the decade, however, the outcomes of the MacArthur Violence Risk Assessment Study produced conflicting data in the literature about the correlations between violence and mental illness (or particular symptoms). This 3-site study involved nearly 1000 psychiatric patients discharged into the community and followed for 1 year, along with a community control sample at 1 site. The MacArthur study was designed to specifically address weaknesses in past research efforts, such as the usual retrospective design and reliance on self-report data. In the first report, in 1998, Steadman and others concluded that the prevalence of violence among discharged psychiatric patients was not significantly different from the rate of violence among community control subjects when neither group demonstrated symptoms of substance abuse (30). The MacArthur data were later used to investigate the reported correlation of TCO symptoms with violence. In 2000 Appelbaum and others (31) reported that Link and Steuve’s findings (27) were replicated when patients’ self-reports about TCO symptoms were used in the data analysis. However, the MacArthur study design approached with caution the labelling of reported symptoms as “delusional” and had trained interviewers rate whether particular responses from subjects were truly delusional. When the interviewers’ ratings were used in the data analysis, no significant relation could be found between TCO symptoms and violence. In fact, a further analysis, conducted to create a classification tree approach to the identification of risk, found that TCO symptoms and a diagnosis of schizophrenia both correlated negatively with violence (32). In the MacArthur group’s revised Iterative Classification Tree (ICT), questions about a diagnosis of major mental disorder appear only twice—both times leading to a classification of low risk (33). (The ICT is described more fully later in this paper.) Swanson and colleagues obtained contrary results from a 1-year posthospitalization follow-up study of patients subject to outpatient commitment in North Carolina (34). In 2000 these researchers reported their findings that paranoid symptoms and TCO symptoms (among other factors, such as low levels of social support and low Global Assessment of Functioning [GAF] score) were both significantly associated with violence during the study period (34). Interestingly, in a 1999 report of North Carolina patients subject to outpatient commitment during a 4-month period preceding hospitalization, Swanson and others did not find an association between violence and paranoia, threat symptoms, schizophrenia-related diagnosis, or GAF score (35). However, another method of investigation—birth cohort studies—continues to yield significant odds ratios (ORs) for serious mental disorders and violence. The OR represents the increased risk, compared with a control group. In a 30-year study of violence in Sweden, Hodgins reported an OR of 4 (that is, a fourfold increase in risk, compared with control subjects) for major mental disorder among male subjects and an OR of 27 for major mental disorder among female subjects (36). In a 26-year study in Finland, Tiihonen and others reported an OR of 7 for male subjects with schizophrenia (37). In a 44-year study in Denmark, Brennan and others found an OR of 4.6 for male subjects with schizophrenia and an OR of 23 for female subjects with schizophrenia (38). In a 21-year New Zealand cohort study, Arsenault and others reported an OR of 2.5 for subjects with schizophrenia spectrum disorders (39). A recently reported Australian study by Wallace and colleagues compared patients with schizophrenia with a matched sample of community individuals in terms of criminal convictions over a 25-year period (40). Substance abuse among the schizophrenia patients increased the likelihood of conviction sixteenfold: 30% of male subjects with both schizophrenia and substance abuse had been convicted of a violent offense. Among all subgroups of schizophrenia patients, the rate of conviction, particularly for violent offenses, was between 3.6 and 6.6 times higher than the rate for community subjects. The authors opine that the findings suggest 6% to 11% of violent offenses in the community may be attributable to schizophrenia patients. Substance Abuse and Violence Multiple studies have revealed a significant correlation between violence and substance use, either alone or combined with other mental disorders. The Swanson and others 1990 ECA data study discussed above was one of the earliest sources of information that substance abuse is a more important variable in risk for violence than any of the major mental disorders (25). Fulwiler and colleagues studied patients in a Boston assertive community treatment program, exploring differences between a violent subgroup and a nonviolent subgroup (41). Although the presence of a major mental disorder did not distinguish the groups, substance abuse—either alone or combined with a major mental disorder—significantly increased the likelihood of a patient’s being in the violent group (41). Similarly, the MacArthur study findings also revealed no effect of mental disorder on violence when patients were not using substances but found significant increases in violence among both patients and community control subjects when they were using substances (30). Swartz and others reported on the North Carolina outpatient commitment data for patients in the 4-month period preceding hospitalization (42). They found that the combination of substance abuse and medication noncompliance strongly predicted serious violent behaviour. The other North Carolina outpatient commitment studies described above also found an association between substance abuse and violence, both prior to hospitalization and in the 1-year period following hospitalization (34,35). Other Factors and Violence In 2002 Swanson and others reported on a multistate sample of public psychiatric patients with psychotic or major mood disorders (43). They identified several environmental factors that were significantly associated with violence during the prior 1-year period, including homelessness and witnessing or experiencing violence. Substance abuse was also significantly associated with violence, as were diagnoses of mood disorder and posttraumatic stress disorder. Other sociodemographic and environmental factors have been found to have significant correlations with violence. The Swanson and others ECA study found that male sex, young age, and low socioeconomic status were significantly correlated with violence (25). The Link and others epidemiologic study found the same results for sex and age and also reported the significant correlation of education level with violence (26). These authors also noted that the risk related to mental illness was less than that attributable to sex or age and roughly equivalent to a 4- to 5-year difference in education. Silver and colleagues attempted to begin an understanding of “ecological” variables associated with violence among individuals with mental illness (44). They found that patients discharged to neighborhoods of “concentrated poverty” had a risk of violence 2.7 times greater than that of similar patients not discharged into such neighborhoods. Many researchers have been careful to point out that the relative contribution of mental illness to the overall rate of violence in society is quite small (23,26,27,45). Link and Steuve specifically caution that most members of even high-risk groups are not violent (27). This point was amply illustrated in the ECA data for factors other than mental health: 84% of those in the highest-risk group (that is, young men in the group with lowest socioeconomic status) were not violent. According to the same data, 87% to 90% of the individuals with psychiatric disorders were not violent (25). The point remains valid even in research specifically designed to accurately predict violent behaviour among psychiatric populations. According to the MacArthur group’s sophisticated ICT methodology, less than one-half (45%) of the individuals operationally identified as high-risk are actually violent (33). Summary Several principles can be extracted from these studies. Substance abuse is a consistently reported risk factor for violence, both alone and combined with other mental health factors. Violence is significantly correlated with various sociodemographic and environmental factors, while the contribution of mental illness is relatively small. Conflicting research findings indicate that the relation between psychosis and violence is unclear. Contemporary Research on Prediction AccuracyWith these advances in knowledge about the correlates of violence, we would expect the accuracy of risk assessment methodologies to have improved since Monahan’s 1981 review of the literature. Indeed, more recent reviews of the state of the art and science of risk assessment have concluded that there is reason for optimism (23,24). In 1997 Monahan noted that the available research demonstrated that “clinical predictions of violence have more than chance validity” (23). Several examples of such studies are worth describing. Clinical Predictions In 1993 Lidz and others reported a study of nearly 2000 university hospital psychiatric emergency department patients for whom clinical staff predicted violence over the next 6 months (46). Various sources were used to follow for violent incidents. The researchers found that prediction accuracy was significantly above chance, with sensitivity equal to 60% and specificity equal to 58%. (Sensitivity here refers to the percentage of individuals who ultimately became violent identified as such by the clinical staff. Specificity refers to the percentage of individuals who did not become violent so identified by staff.) In addition, the patients predicted to become violent committed more serious acts of violence, compared with patients not predicted to become violent. The predictive accuracy was sex-dependent, however, in that the clinicians significantly underestimated the rates of violence among female patients and thus did not achieve above-chance predictions for this population. A further interesting component of this study was that the authors compared the accuracy of the clinical predictions with predictions that could have been made solely on history of violence. If history alone had been used as a predictor, the sensitivity would have been stronger (69%), but the specificity would have been weaker (48%), than that produced by clinical prediction. Clinical prediction was better than chance even when variables such as history, age, and sex were controlled. In 1994 Mossman reanalyzed 58 data sets from 44 published studies of violence prediction, using receiver operating characteristics as a statistical methodology not hampered by problems of base rate or preference for Type I or Type II errors (47). He found that, contrary to the widely accepted view that short-term predictions of violence were more accurate than long-term predictions, accuracy of prediction across time periods did not differ. In 47 of the data sets, accuracy was significantly better than chance, with an area under the curve (AUC) = 0.73. (AUC is calculated from a plot of the true-positive rate vs the false-positive rate. An AUC = 0.5 represents chance predictive validity, while an AUC = 1.0 represents perfect predictive validity.) He also found that studies after 1986 reported more accurate predictions than those published prior to that year, although he noted that this could be the result of reporting bias rather than of improved validity. Finally, he discovered that, among these studies, past behaviour was a better indicator of future violence than was clinical judgment. Mulvey and Lidz advanced the inquiry on prediction accuracy by posing the question of whether clinicians had any ability to predict when and under what circumstances patients would become violent (48). Reported in 1998, this study also involved patients drawn from the psychiatric emergency department. Clinicians were asked to predict which patients would become violent, under what conditions the violence would occur, and what characteristics the violence would take. The researchers found that clinicians did reasonably well in their predictions about place, target, severity of violence, and the involvement of alcohol in the violence. The clinicians significantly overestimated the roles of medication noncompliance and drug use as conditional factors. They also tended to focus their predictions on clinical conditions (that is, problems they might be able to influence or at least monitor) rather than on social or environmental factors. Skeem, Mulvey, and Lidz ventured further into the assessment of conditions in predicting violence by assessing the accuracy of predictions predicated on the condition itself. In particular, they studied the accuracy of clinicians’ predictions that patients would be violent if they were using alcohol. These authors found that clinicians were moderately accurate in predicting who would be violent, that patients who drank were more likely to be violent, and that clinicians were more likely to predict violence in patients who drank heavily. However, clinicians were not able to discriminate well between drinking patients who do and do not become violent (49). Several studies have also looked at the accuracy of violence predictions on inpatient units. McNeil and Binder studied 149 involuntary civil patients to assess physician and nurse accuracy in predicting violence during the first 7 days of hospitalization (50). They found that nurses overpredicted actual assaults but quite accurately predicted the risk of any aggressive acts occurring. Physicians accurately predicted low and medium risks for violence but overpredicted aggression among the group of patients about whom they were most concerned. In a subsequent inpatient study, McNeil and Binder created a 5-item screening actuarial instrument based on items correlating with inpatient violence in a previous study (51). The presence of 3 or more factors was deemed high risk. This simple instrument yielded better-than-chance discrimination between aggressive and nonaggressive patients, with 57% sensitivity and 70% specificity. In a Madrid study of inpatients with schizophrenia or schizoaffective disorder, Arango and others reported that inpatient violence was not related to sociodemographic variables, psychiatric history, or substance use in the preceding week (52). Rather, it was best explained by violence in the week preceding hospitalization, by general psychopathology score on the Positive and Negative Syndrome Scale, and by poor insight into psychotic symptoms—with the last being the single best predictor. By creating an actuarial prediction model using these 3 factors, the researchers were able to correctly identify whether 84% of the sample would exhibit violence during the hospitalization, with a positive predictive power of 80%. In 1999 Hoptman and others reported a study of the clinical prediction of violence among male patients admitted to a maximum-security psychiatric facility. While the clinicians correctly classified 71% of the sample (sensitivity, 54%; specificity, 79%), the characteristics on which they based their predictions did not match characteristics subsequently found to be associated with violence in the study population: the clinicians did not use factors that actually turned out to be predictive, such as a dual diagnosis of schizophrenia–substance use and thought disorder, but did use factors that were not found to be predictive, such as race, ability to follow ward routine, and arrest history. The factors used by clinicians that correctly predicted violence included transfer from a civil hospital owing to violence or dangerousness, age, childhood physical abuse, and temper (53). In a university hospital study in Israel, Haim and others compared the ability of doctors and nurses to predict inpatient violence among a sample population of 308 consecutively admitted patients. The groups performed equally well and used similar factors in their predictions (such as previous acquaintance with the patient, patients’ threats, verbal aggression, and property-damaging behaviours). The psychiatrists correctly classified the patients in 82% of the cases; however, this was mostly owing to the correct prediction of nonviolence, since the specificity was 88%, compared with sensitivity of only 37%. The context for correctly predicting nonviolence was a 10.7% base rate of violence in the sample (54). Actuarial Predictions During this contemporary era of risk research, there has also been substantial progress in actuarial prediction of violence. Of course, proponents of actuarial instruments have been arguing their superiority since the 1940s and 1950s (55,56). Various instruments have been developed, and their validity in predicting violence has empirical support. The Psychopathy Checklist-Revised (PCL-R) was originally developed as a research tool to study antisocial personality disorder (57). However, it has been widely used to predict various types of violence. The results show strong correlations with criminal recidivism, violence, and sexual violence. Compared with other structured tools, the PCL-R may be the best predictor of future violence (58). Part of its strength is that it identifies traits that are stable over time. Thus it predicts lifelong risk rather than imminent risk and is, like other actuarial instruments, insensitive to changes in clinical state. The Violence Risk Appraisal Guide (VRAG) has 12 variables, of which the PCL-R score is the most heavily weighted (59). Other variables include mostly historical information but also include several diagnostic variables—personality disorder, alcohol abuse, and schizophrenia (the last having an inverse relation to risk of violence). The instrument was developed on a sample of violent offenders and has been tested with sex offenders, maximum-security prison populations, and forensic patients. In the latter study sample, Harris and others reported that, on 5-year follow-up, the VRAG demonstrated an AUC = 0.80, compared with a proxy of clinical prediction that had an AUC = 0.62 (60). The VRAG authors initially argued that clinicians might wish to consider adjustments to the empirically derived risk assessment, based on dynamic variables not included in the VRAG, such as progress in treatment, family support, access to weapons, change in procriminal attitudes, and amount and quality of supervision in the environment (59). Subsequently, however, the authors argued that the VRAG’s superior accuracy in predicting violence demands that it be used to assess risk without modification and in preference to clinical methods and that clinical skill should be reserved for gathering interview data and scoring the instrument (60). Another instrument that is being increasingly studied is the Historical/Clinical/Risk Management 20-item (HCR-20) (61). The instrument incorporates 10 historical items, 5 clinical items, and 5 risk management items. It has been noted that the presence of the clinical items offers potential for incorporating the effects of treatment on risk levels (56,58). Several studies have demonstrated the utility of the HCR-20 in predicting violence in various settings (58,62–64). Finally, a recent goal of the MacArthur Study group has been to incorporate an exhaustive array of possible risk factors into a comprehensive risk assessment instrument, the ICT. The MacArthur group developed this tool by culling 134 possible risk factors from the available literature and incorporating them into an empirical analysis of their relevance to violence risk in the 939-patient MacArthur study population (32). In a subsequent revision of the ICT (that restricted it to 106 items that were generally available in clinical records), 72.6% of the subjects could be classified as either high or low risk (33). The risk factors were divided across 4 domains: dispositional or personal, historical or developmental, contextual or situational, and clinical or symptom. Empirical analysis demonstrated that psychopathy had the highest correlation with violence, followed by a history of prior arrests. Although substance use and anger were the next-highest correlates on the list, no other clinical variables showed high correlation with violence in this group of discharged patients. The 2 items next on the list did not even involve the subjects but, rather, their fathers’ use of drugs and histories of arrest. Although 2 domains advantageously comprised potentially dynamic variables, these variables did not fall out as empirically useful to the categorization of high or low risk. Thus the ICT represents a largely static actuarial approach. Summary Several principles can be extracted from these studies. Clinical assessments of risk achieve better-than-chance accuracy, although clinicians may not be particularly good at identifying specific conditions under which violence will occur. Actuarial risk assessments achieve statistically superior accuracy owing to the stability of historical factors in indicating lifelong risk. However, actuarial methods are inherently insensitive to change and cannot inform clinicians’ assessment of treatment progress. ConclusionsFaced with this substantial but complicated body of knowledge about violence risk factors and predictive accuracy, clinicians will be left with at least the following important pragmatic questions. What can be concluded about risk assessment in various clinical situations? Is the available knowledge clinically useful? Can it be fairly or accurately applied to individual patients? The research on violence has not produced a clearly uniform picture of the most important mental health variables associated with risk of violent behaviour. The one exception is the factor of substance abuse, which has been universally associated with a significantly increased risk of violence, far surpassing the contribution of serious mental illness. Thus it is clear that any risk-focused clinical management plan should always target substance abuse for significant intervention efforts in clinical settings. Another area of general agreement in the literature is that variables other than mental health contribute more significantly to the overall rate of violence in the population than do mental health variables (23,26,27,45). Thus sociodemographic factors such as young age, male sex, and low socioeconomic status (25,26), as well as environmental influences such as residing in a neighborhood of concentrated poverty (44), create a baseline risk for violence that will confound clinicians’ attempts to analyze and ameliorate risk. For variables over which we have more control, namely, the treatment of various mental disorders and symptom clusters, the evidence regarding the association with violence is mixed. Most studies can be summarized as supporting the idea that serious mental illness does moderately increase the risk for violence (25–29,34,36–40). Epidemiologic and birth-cohort studies, in particular, tend to support this conclusion. Other studies, however, including the noteworthy MacArthur studies, support a conclusion that mental health variables do not discriminate well between subgroups of violent and nonviolent individuals (30–32,35,41). In fact, the MacArthur database supports a conclusion that a diagnosis of schizophrenia and the presence of TCO symptoms are both negatively correlated with violence (32). How are clinicians to make sense of a literature supporting conclusions that schizophrenia is negatively correlated with violence and yet increases the risk of violence? One answer may be McNeil and colleagues’ observation that clinical factors seem to be most relevant for acutely ill individuals, whereas historical data may be most relevant for treated patients and for assessment of long-term risk (63). This may, in fact, explain the clinically counterintuitive findings of the MacArthur studies (in which the subjects were all patients released from inpatient treatment by the natural decisions of the treating clinicians). This inconsistency in the research data may also simply represent evidence that psychiatric treatment effectively reduces risk for violence associated with clinical symptoms. Nonetheless, this presents a good news–bad news situation for clinicians. It would be good news to know that the important risk factors for violence are closely related to the important clinical factors bringing patients to acute care. We are relatively well equipped scientifically to manage the acute manifestations of serious mental illness, even if politically and economically hindered at times in our ability to deliver appropriate and fully effective care. The bad news of such an observation is that the greater accuracy of historical factors (and thus actuarial models) in predicting long-term risk might mean that clinicians will be held responsible for factors that they cannot mitigate by even their most careful efforts. It also means that patients are at risk of being confined more extensively than can be justified by the dynamic clinical improvement presumed in the societal quid pro quo of involuntary commitment. We do not yet know how clinical treatment might reduce actuarially determined high risk or even whether it can be reduced. Lindquist and Skipworth refer to this phenomenon as “guilt by statistical association” (65). They argue that, rather than assessing their historically based risk, we should instead ask more often about the rehabilitative tasks to be undertaken with our patients. Among other imperatives he outlines, Mullen argues that the ethical conduct of risk assessment in clinical practice requires mental health variables to be prominent and directly relevant to the probability of violent behaviour in each person and situation evaluated (66). In his view, clinical risk assessments should be motivated “primarily by the intention to provide the patient with better treatment and care” (66, p 2068). Dvoskin and Heilbrun recommend a model in which the results of actuarial risk assessment would be communicated, with appropriate description of the strengths and limitations of the methodology applied, when the “best available prediction of violence risk” is desired (56, p 9). However, they also recognize that actuarial risk is unlikely to be reduced by a course of clinical treatment. Thus they recommend a “risk management approach” to clinical work: significant clinical risk factors are to be identified and targeted for intervention, and the outcomes of clinical treatment are to be fed into a reassessment of the risk factors. Elsewhere, Heilbrun and colleagues have demonstrated that such a risk management approach is the method preferred by mental health experts for communicating risk when dealing with high-risk cases (67). Dvoskin has also described this thorough clinical approach as the “anamnestic method” (68), emphasizing the ethnographic element of understanding a person’s history, patterns of behaviour, and response to situations, as well as the assessment of the skills needed to avoid violent response to future triggering situations. Mullen has offered a similar model designed to manage imminent risk and provide a long-term strategy to minimize harm (66). To arrive at an assessment of risk according to this model, clinicians assess preexisting vulnerabilities, protective and aggravating influences in the social and interpersonal environment, mental disorder and substance abuse, and the presence of situational triggers. Remedial action is directed to those factors conducive to escalating risk, with the patient’s needs at the centre of the clinician’s focus. Mullen elaborates on each factor and its assessment. These risk management strategies probably represent the best approach clinicians can take in response to the available data and the current demands on clinical practice. Two other factors, however, must be highlighted as cause for our continued humility in matters of risk assessment and management. The first is that we know relatively little about the interplay between clinical factors and contextual and environmental issues (44,48,49). We have little control over the environments in which our patients live or over the socioeconomic and political challenges they face. Although we know something about the factors correlated with violence, this knowledge does not help us make uniformly valid predictions about the conditions under which our patients might become violent in the future (48). The second cause for humility is that the low base rate for violence and the weak contribution of mental health variables to violence in society combine to make all our risk assessment activities (including those conducted via actuarial study, 69) open to high rates of false-positive error. In every high-risk population identified, nonviolent individuals outnumber violent individuals. In forensic populations, patients can never shed their historical risk factors. Thus any approach to risk assessment that relies exclusively (or nearly so) on these factors will deprive patients and their clinicians of hope for recovery. The pursuit of scientific accuracy at the expense of hope seems to run counter to the essence of the healing professions, and clinicians are therefore cautioned to follow the literature with these limitations clearly in mind. AcknowledgementsThe authors thank Dr Ezra Griffith, Dr Howard Zonana, Dr Alec Buchanan, Dr Marsha Kline Pruett, Dr Adrienne Roth, and Ms Josephine Buchanan for their comments on an earlier draft of this paper. References1. Bush GHW. Presidential Proclamation 6158, July 17, 1990. Washington (DC): Office of the Federal Register; 1990. 2. Phelan JC, Link BG. The growing belief that people with mental illnesses are violent: the role of the dangerousness criterion for civil commitment. Soc Psychiatry Psychiatr Epidemiol 1998;33(Suppl 1):S7–S12. 3. Norko MA. Commentary: Dangerousness—a failed paradigm for clinical practice and service delivery. J Am Acad Psychiatry Law 2000;28:282–9. 4. Brakel SJ, Parry J, Weiner BA. The mentally disabled and the law. Chicago (IL): The American Bar Association; 1985. p 178. 5. Baxstrom v Herold, 383 US 107 (1966). 6. Lake v Cameron, 364 F2d 657 (DC Cir 1966). 7. Hawks v Lazaro, 157 W Va 417, 202 SE 2d, 109 (1974). 8. O’Connor v Donaldson, 422 US 563, 95 SCt 2486 (1975). 9. Tarasoff v Regents of the University of California, 17 Cal 3d 425 (1976). 10. Lipari v Sears Roebuck, 497 F Supp 185 (D Neb 1980). 11. Peck v Counseling Service of Addison County, 499 A2d 422 (1985). 12. Estelle v Smith, 451 US 454 (1980). 13. Barefoot v Estelle, 463 US 880 (1982). 14. Jones v United States, 463, US 354 (1983). 15. Monahan J. The clinical prediction of violence. DHHS Publication No (ADM) 81–921. Rockville (MD): US Department of Health and Human Services; 1981 16. Diamond B. The psychiatric prediction of dangerousness. University of Pennsylvania Law Review 1974;123:439–52. 17. Shah S. Dangerousness: a paradigm for exploring some issues in law and psychology. Am Psychol 1978;33:224–38. 18. Stone AA. Comment on Peszke MA—Is dangerous an issue for physicians in emergency commitment? Am J Psychiatry 1975;132:829–31. 19. Roth LH. A commitment law for patients, doctors, and lawyers. Am J Psychiatry 1979;136:1121–7. 20. American Psychiatric Association. Guidelines for legislation on the psychiatric hospitalization of adults. Am J Psychiatry 1983;140:672–9. 21. Lansing AE, Lyons JS, Martens LC, O’Mahoney MT, Miller SI, Obolsky A. The treatment of dangerous patients in managed care—psychiatric hospital utilization and outcome. Gen Hosp Psychiatry 1997;19:112–8. 22. Kansas v Hendricks, 117 S Ct 2072 (1997). 23. Monahan J. Clinical and actuarial predictions of violence. In: Faigman D, Kaye D, Saks M, Sanders J., editors. Modern scientific evidence; the law and science of expert testimony. St Paul (MN): West Publishing Company; 1997. 24. Steadman HJ. From dangerousness to risk assessment of community violence: taking stock at the turn of the century. J Am Acad Psychiatry Law 2000;28:265–71. 25. Swanson JW, Holzer CE, Gangu VK, Jono RT. Violence and psychiatric disorder in the community: evidence from the Epidemiologic Catchment Area surveys. Hosp Community Psychiatry 1990;41:761–70. 26. Link BG, Andrews H, Cullen FT. The violent and illegal behavior of mental patients reconsidered. Am Sociol Rev 1992;57:275–92. 27. Link BG, Stueve A. Psychotic symptoms and the violent/illegal behavior of mental patients compared to community controls. In: Monahan J, Steadman H, editors. Violence and mental disorder: developments in risk assessment. Chicago (IL): University of Chicago Press; 1994. 28. Swanson JW, Borum R, Swartz M, Monahan J. Psychotic symptoms and disorders and the risk of violent behaviour in the community. Criminal Behavior and Mental Health 1996;6:317–38. 29. Link BG, Stueve A, Phelan J. Psychotic symptoms and violent behaviors: probing the components of “threat/control-override” symptoms. Soc Psychiatry Psychiatr Epidemiol 1998;33:S55–S60. 30. Steadman HJ, Mulvey EP, Monahan J, Robbins PC, Appelbaum PS, Grisso T, and others. Violence by people discharged from acute psychiatric inpatient facilities and by others in the same neighborhoods. Arch Gen Psychiatry 1998;55:393–401. 31. Appelbaum PS, Robbins PC, Monahan J. Violence and delusions: data from the MacArthur Violence Risk Assessment Study. Am J Psychiatry 2000;157:566–72. 32. Steadman HJ, Silver E, Monahan J, Appelbaum PS, Robbins PC, Mulvey EP, and others. A classification tree approach to the development of actuarial violence risk assessment tools. Law Hum Behav 2000;24:83–100. 33. Monahan J, Steadman HJ, Robbins PC, Silver E, Appelbaum PS, Grisso T, and others. Developing a clinically useful actuarial tool for assessing violence risk. Br J Psychiatry 2000;176:312–9. 34. Swanson JW, Swartz MS, Wagner HR, Burns BJ, Borum R, Hiday VA. Involuntary out-patient commitment and reduction of violent behaviour in persons with severe mental illness. Br J Psychiatry 2000;176:324–31. 35. Swanson JW, Borum R, Swartz MS, Hiday V. Violent behavior preceding hospitalization among persons with severe mental illness. Law Hum Behav 1999;23:185–204. 36. Hodgins S. Mental disorder, intellectual deficiency, and crime: evidence from a birth cohort. Arch Gen Psychiatry 1992;49:476–83. 37. Tiihonen J, Isohanni M, Rasanen P, Koiranen M, Moring J. Specific major mental disorders and criminality; a 26-year prospective study of the 1966 northern Finland birth cohort. Am J Psychiatry 1997;154:840–5. 38. Brennan PA, Mednick SA, Hodgins S. Major mental disorders and criminal violence in a Danish birth cohort. Arch Gen Psych 2000;57:494–500. 39. Arsenault L, Moffitt Te, Caspi A, Taylor PJ, Silva PA. Mental disorders and violence in a total birth cohort. Arch Gen Psychiatry 2000;57:979–86. 40. Wallace C, Mullen PE, Burgess P. Criminal offending in schizophrenia over a 25-year period marked by deinstitutionalization and increasing prevalence of comorbid substance use disorders. Am J Psychiatry 2004;161:716–27. 41. Fulwiler C, Grossman H, Frobes C, Ruthazer R. Early-onset substance abuse and community violence by outpatients with chronic mental illness. Psychiatr Serv 1997;48:1181–5. 42. Swartz MS, Swanson JW, Hiday VA, Borum R, Wagner R, Burns BJ. Taking the wrong drugs; the role of substance abuse and medication noncompliance in violence among severely mentally ill individuals. Soc Psychiatry Psychiatr Epidemiol 1998;33:S75–S80. 43. Swanson JW, Swartz MS, Essock SM, Osher FC, Wagner HR, Goodman LA, and others. The social-environmental context of violent behavior in persons treated for severe mental illness. Am J Public Health 2002;92:1523–31. 44. Silver E, Mulvey EP, Monahan J. Assessing violence risk among discharged psychiatric patients: toward an ecological approach. Law Hum Behav 1999;23:237–55. 45. Walsh E, Buchanan A, Fahy T. Violence and schizophrenia: examining the evidence. Br J Psychiatry 2001;180:490–5. 46. Lidz CW, Mulvey EP, Gardner W. The accuracy of predictions of violence to others. JAMA 1993;269:1007–11. 47. Mossman D. Assessing predictions of violence: being accurate about accuracy. J Consult Clin Psychol 1994;62:783–92. 48. Mulvey EP, Lidz CW. Clinical prediction of violence as a conditional judgment. Soc Psychiatry Psychiatr Epidemiol 1998;33:S107–S113. 49. Skeem JL, Mulvey EP, Lidz CW. Building mental health professionals’ decisional models into tests of predictive validity: the accuracy of contextualized predictions of violence. Law Hum Behav 2000;24;607–28. 50. McNeil DE, Binder RL. Clinical assessment of the risk of violence among psychiatric inpatients. Am J Psychiatry 1991;148:1317–21. 51. McNeil DE, Binder RL. Screening for risk of inpatient violence: validation of an actuarial tool. Law Hum Behav 1994;18:579–86. 52. Arango C, Barba AC, Gonzalez-Salvador T, Ordonez AC. Violence in inpatients with schizophrenia; a prospective study. Schizophr Bull 1999;25:493–503. 53. Hoptman MJ, Yates, KF, Patalinjug MB, Wack RC, Convit A. Clinical prediction of assaultive behavior among male psychiatric patients at a maximum-security forensic facility. Psychiatr Serv 1999;50:1461–6. 54. Haim R, Rabinowitz J, Lereya J, Fennig S. Predictions made by psychiatrists and psychiatric nurses of violence by patients. Psychiatr Serv 2002;53:622–4. 55. Buchanan A. Risk and dangerousness. Psychol Med 1999;29:465–73. 56. Dvoskin JA, Heilbrun K. Risk assessment and release decision-making: toward resolving the great debate. J Am Acad Psychiatry Law 2001;29:6–10. 57. Hare RD. Manual for the Hare-Psychopathy Checklist-Revised. Toronto: Multi-Health Systems; 1991. 58. Dolan M, Doyle M. Violence risk prediction. Br J Psychiatry 2000;177:303–11. 59. Harris GT, Rice ME. Risk appraisal and management of violent behavior. Psychiatr Serv 1997;48:1168–76. 60. Harris GT, Rice ME, Cormier CA. Prospective replication of the Violence Risk Appraisal Guide in predicting violent recidivism among forensic patients. Law Hum Behav 2002;26:377–94. 61. Webster CD, Douglas KS, Eaves D, Hart SD. HCR-20: assessing risk of violence, version 2. Burnaby (BC): Simon Fraser University; 1997. 62. Gray NS, Hill C, McGliesh A, Timmons D, MacCulloch MJ, Snowden RJ. Prediction of violence and self-harm in mentally disordered offenders: a prospective study of the efficacy of HCR-20, PCL-R, and psychiatric symptomatology. J Consult Clin Psychol 2003;71:443–51. 63. McNeil DE, Gregory AL, Lam JN, Binder RL, Sullivan GR. Utility of decision support tools for assessing acute risk of violence. J Consult Clin Psychol 2003;71:945–53. 64. Douglas KS, Ogloff JRP, Hart SD. Evaluation of a model of violence risk assessment among forensic psychiatric patients. Psychiatr Serv 2003;54:1372–9. 65. Lindqvist P, Skipworth J. Evidence-based rehabilitation in forensic psychiatry. Br J Psychiatry 2000;176:320–3. 66. Mullen PE. Dangerousness, risk and the prediction of probability. In: Gelder MG, López-Ibor JJ, Andreasen NC, editors. New Oxford textbook of psychiatry. Oxford (UK): Oxford University Press; 2001. 67. Heilbrun K, O’Neill ML, Strohman LK, Bowman Q, Philipson J. Expert approaches to communicating violence risk. Law Hum Behav 2000;24:137–48. 68. Dvoskin J. Knowledge is not power—knowledge is obligation. J Am Acad Psychiatry Law 2002;30:533–40. 69. Freedman D. False prediction of future dangerousness: error rates and psychopathy checklist-revised. J Am Acad Psychiatry Law 2001;29:89–95. Author(s)Manuscript received and accepted November 2004. 1. Associate Clinical Professor, Department of Psychiatry, Yale University School of Medicine, New Haven, Connecticut; Chief of Forensic Services, Whiting Forensic Division of Connecticut Valley Hospital, Middletown, Connecticut. 2. Assistant Clinical Professor, Department of Psychiatry, Yale University School of Medicine, New Haven , Connecticut; Director, New Haven Diversion Project and Associate Director, New Haven Office of Court Evaluations, New Haven, Connecticut. Address for correspondence: Dr MA Norko, Connecticut Mental Health Center, Law and Psychiatry Division, 34 Park Street, New Haven, CT 06519 e-mail: michael.norko@yale.edu
1 | 2
|
||||||
