Proximal Risk Factors for Short-Term Community Violence Among Adults With Mental Illnesses

Abstract

Objective

This study examined the role of static indicators and proximal, clinically relevant indicators in the prediction of short-term community violence in a large, heterogeneous sample of adults with mental illnesses.

Methods

Data were pooled from five studies of adults with mental illnesses (N=4,484). Follow-up data were available for 2,579 participants. A hierarchical linear regression assessed the incremental validity of a series of variable clusters in the prediction of violence risk at six months: static characteristics (age, sex, race-ethnicity, and primary diagnosis), substance use (alcohol use and drug use at baseline), clinical functioning (psychiatric symptoms at baseline and recent hospitalization), recent violence, and recent victimization.

Results

Results demonstrated improved prediction with each step of the model, indicating that proximal indicators contributed to the prediction of short-term community violence above and beyond static characteristics. When all variables were entered, current alcohol use, recent violence, and recent victimization were positive predictors of subsequent violence, even after the analysis controlled for participant characteristics.

Conclusions

This study provides empirical evidence for three proximal, clinically relevant indicators in the assessment and management of short-term violence risk among adults with mental illnesses: current alcohol use, recent violence, and recent victimization. Consideration of these indicators in clinical practice may assist in the identification of adults with mental illnesses who are at heightened risk of short-term community violence.

In light of recent tragedies, including the Newtown, Connecticut, and Aurora, Colorado, shootings, increased attention has been focused on the elevated risk of violence perpetration among adults with mental illnesses, compared with the general population (1–4), and with the role of mental health professionals in preventing such violence (5,6). Although clinicians are tasked with the identification of risk factors associated with community violence, research efforts have focused primarily on static factors (for example, age, sex, and race-ethnicity) with relatively limited clinical utility (7). When clinically relevant factors are considered, they often are distal in nature (for example, childhood victimization experiences and a history of substance use) and may not be representative of current functioning and, therefore, of the current risk of violence. However, some evidence suggests that proximal factors—that is, factors that are currently present or that occurred in the recent past—may play an important role in the assessment of violence risk, particularly vis-à-vis the prediction of short-term violence (8,9). Four proximal indicators stand out as potentially relevant to the clinical prediction and prevention of short-term community violence among adults with mental illnesses: current substance use, current psychiatric symptoms, recent violence, and recent victimization.

Alcohol and drug use are widely recognized as robust correlates of violence perpetration among adults with mental illnesses (4,10,11). However, research-based and clinical assessments of violence risk often rely on a diagnosis of a substance use disorder or self-reported history of substance use problems, rather than on current levels of use (12,13). In addition, alcohol and drug use are often considered together under the umbrella of substance use, although past research differs as to which evinces a stronger relationship with violence (14,15). Thus information on prior substance use problems, although predictive of long-term violence, provides little in the way of information regarding current functioning, risk of violence in the short-term, and treatment needs.

As with substance use, having a psychiatric diagnosis, compared with having no diagnosis, is associated with an increased risk of violence (4). However, evidence regarding the role of psychiatric symptoms in relation to violence is mixed. Prior reviews have not reached a consensus regarding the strength or even direction of the association between psychiatric symptoms and violence (16–18). These equivocal findings may reflect, in part, heterogeneity in operational definitions; studies vary in their use of total scores, individual symptoms, and latent factors (10,19). Furthermore, most research on the association between psychiatric symptoms and violence has focused on long-term violence (that is, more than 12 months after assessment) (11,16). Few studies have examined the associations between current psychiatric symptomatology and violence over the coming months. As a result, psychiatric symptoms are often supplanted in assessments of violence risk by distal measures of psychiatric diagnoses or insensitive proximal measures of exacerbated symptomatology (for example, recent hospitalization).

In addition to being perpetrators of violence, adults with mental illnesses are also victims of violence—and at rates higher than those in the general population (1,20,21). Both a history of violence and a history of victimization have been identified as robust correlates of long-term violence (11,22,23), although the latter is attended to less frequently in violence risk assessment. When a history of victimization is considered, the focus is typically on distal experiences (for example, a history of childhood abuse) (24). Although childhood and adult forms of victimization are related among adults with mental illnesses (25), the experiences are qualitatively distinct and thus may evince different associations with short-term violence. Indeed, a recent study of 167 justice-involved adults with mental illnesses showed that recent violence and victimization predicted violence over a 12-month period, above and beyond static and distal factors, including childhood physical abuse (24). However, generalizability of findings from this study is limited by the small sample and the authors’ decision to collapse six- and 12-month assessments of violence.

This study sought to better understand the role of proximal indicators in the prediction of short-term violence through analysis of integrated data. Specifically, we examined current alcohol and drug use, current psychiatric symptomatology, recent hospitalization, recent violence, and recent victimization as predictors of community violence over a six-month period in a large, heterogeneous sample of adults with mental illnesses.

METHODS

Data

Data were pooled from five studies of adults with mental illnesses (N=4,484). The most recent was a study of facilitated psychiatric advance directives (F-PAD study; N=473) (10) that investigated a psychiatric advance directive intervention, with interviews conducted between 2003 and 2007. The MacArthur Violence Risk Assessment Study (MacRisk Study; N=1,136) (2) evaluated violence among psychiatric patients on civil commitments; data were collected through participant and collateral interviews and abstraction from hospital records from 1992 to 1995. The Schizophrenia Care and Assessment Program (SCAP; N=404) (26) examined clinical, functional, and service outcomes for adults with schizophrenia from 1997 to 2002. The MacArthur study of mandated community treatment (MacMandate; N=1,011) (27) assessed treatment leverage among psychiatric out-patients between 2002 and 2003 (1). The Clinical Anti-psychotic Trials of Intervention Effectiveness (CATIE; N=1,460) (28) examined the effectiveness of antipsychotic medications among adults with schizophrenia between 2001 and 2004.

The protocol for the study reported here was approved by institutional review boards from North Carolina State University, RTI International, and Arizona State University. All participants gave written informed consent. Parent studies enrolled a range of participants, from inpatients with exacerbated symptoms to outpatients in partial remission. Together, the characteristics of the sample used for this study approximate those of a usual-care, noninterventional population. Analyses were conducted with data from 2,579 participants available at six-month follow-up (74% of the longitudinal subsample).

Measures

Outcome variable

Prevalence and severity of community violence were assessed in all studies by using the MacArthur Community Violence Screening Instrument (MCVSI) (2). The MCVSI comprises eight questions each for violence and victimization; questions are derived from the Revised Conflict Tactics Scale (29). Specifically, the questions assess pushing, grabbing, or shoving; kicking, biting, or choking; slapping; throwing an object; hitting with a fist or object; sexual assault; threatening with a weapon in hand; and using a weapon. For each item, participants are first asked whether someone did this to them and then whether they did this to someone else. Factor analyses (not presented here but available upon request) showed that the violence and victimization items each mapped onto a unidimensional factor in the current sample. These findings are consistent with prior psychometric evaluations of the MCVSI (20,30).

For the analyses reported here, violence and victimization factor scores were created from unidimensional factor models by using expected a posteriori (EAP) estimates. EAP estimates are calculated as the mean of the posterior-predicted distribution of scores for an individual based on his or her response pattern and the estimated model parameters. The factor scores account for incomplete data and reflect the items to which the participant responded affirmatively, such that a higher score indicates greater prevalence or severity of violence. Because the MacMandate study was cross-sectional in design, participant data were used in the calculation of factor scores but were excluded from predictive analyses. In the MacRisk Study, violence was also assessed three months after the baseline interview; to include these data, we averaged three- and six-month factor scores.

Static predictors

Participant age was measured continuously (in years) at baseline. Sex was coded dichotomously (1, male; 0, female). Race-ethnicity was captured with four variables: white, black, Hispanic, and other. Primary diagnosis was measured with five dummy variables: schizophrenia, bipolar disorder, major depressive disorder, substance use disorder, and other disorder (for example, anxiety disorder). Psychiatric diagnoses were obtained through a combination of clinician diagnoses and medical records abstraction.

Proximal predictors

Current alcohol and drug use were assessed at baseline with multiple measures across studies. Measures included the CAGE questionnaire (31), urine drug screens, self-report, the Alcohol or Drug Use Scales (32), and the Structured Clinical Interview for DSM-IV (33). For both alcohol use and drug use, we harmonized data across studies by creating a three-level indicator (0, abstinence; 1, nonproblematic use; and 2, problematic use).

Our indicators of clinical functioning included current psychiatric symptoms and recent hospitalization. Psychiatric symptoms were assessed at baseline with the Positive and Negative Syndrome Scale (34) in the CATIE and SCAP and with the Brief Psychiatric Rating Scale (35) in the F-PAD and MacRisk studies. Briefly, a factor-analytic cross-validation approach was employed, with exploratory factor analyses conducted on a random subsample of data in which four factors were retained: affect, positive symptoms, negative symptoms, and disorganized cognitive processing. This model was then evaluated and supported by using a confirmatory factor model with the remainder of the data. Factor scores for each latent trait were then created as described above (Van Dorn et al., unpublished manuscript, 2015). Recent hospitalization was measured in all studies and indicated whether a participant was hospitalized in the three months prior to the baseline assessment.

Recent violence and recent victimization referred to the baseline factor scores, created as described above.

Data Analysis

All analyses were conducted with SAS, version 9.4. Descriptive statistics were calculated for all variables, and characteristics were compared between participants who were and were not available at follow-up. For bivariate and multivariable analyses, we used all cases with data at baseline and six-month follow-up (N=2,579). Bivariate associations of predictor variables with violence at six-month follow-up were examined by using a series of one-way analyses of variance and Pearson correlations; post hoc comparisons were conducted by using Bonferroni-adjusted t tests. A hierarchical linear regression analysis assessed the incremental validity of several variable clusters in the prediction of violence risk. Variables were entered into the model in five steps: static characteristics, substance use, clinical functioning, recent violence, and recent victimization. Listwise deletion was used in cases of missing data. We included the parent study as a covariate in multivariable analyses to control for differences across studies.

RESULTS

Participant Characteristics

Frequencies and means for all predictor variables are reported in Table 1. Characteristics of the participants who completed follow-up interviews differed from the full, baseline sample in several ways (Table 1). In general, those present at follow-up exhibited poorer psychosocial functioning at baseline. These factors were included as predictors in subsequent analyses. Of the 2,579 participants available at six-month follow-up, almost two-thirds were male. About half were white, about a third were black, and the remainder identified as Hispanic or other race-ethnicity. Schizophrenia was the most prevalent primary diagnosis, followed by major depression, bipolar disorder, substance use disorder, and other disorder (for example, anxiety disorder). Just over half had been hospitalized within three months of baseline. At baseline, most participants reported abstinence from both alcohol and drugs. Approximately one-quarter of participants reported perpetrating at least one violent act, and about one-third reported experiencing at least one incident of victimization in the six months preceding baseline. At follow-up, 23% of participants reported perpetrating at least one violent act in the past six months. Prevalence rates for each type of act are reported elsewhere (20).

TABLE 1.

Characteristics of a sample of adults with mental illnesses and of a subsample with six-month follow-up data

Characteristic	Full sample (N=4,484)		Follow-up (N=2,579)		pa
	N	%	N	%
Sex					.499
Male	2,681	60	1,554	60
Female	1,800	40	1,025	40
Race-ethnicity					<.001
White	2,300	51	1,353	53
Black	1,687	38	1,036	40
Hispanic	317	7	136	5
Other	172	4	54	2
Primary diagnosis					<.001
Schizophrenia	2,837	64	1,687	65
Bipolar disorder	424	10	218	9
Major depression	824	18	442	17
Substance use disorder	277	6	205	8
Other	106	2	20	1
Recent hospitalization (past 6 months)					<.001
No	2,704	60	1,274	49
Yes	1,776	40	1,304	51
Alcohol use					<.001
Abstinence	2,474	55	1,311	51
Nonproblematic use	858	19	546	21
Problematic use	1,141	26	716	28
Drug use					<.001
Abstinence	3,022	68	1,665	65
Nonproblematic use	476	11	310	12
Problematic use	969	22	594	23
Perpetrated any recent violence (past 6 months)					.594
No	3,443	77	1,977	77
Yes	1,023	23	597	23
Experienced any recent victimization (past 6 months)					.303
No	3,082	69	1,760	68
Yes	1,382	31	812	32
Age (M±SD)	39.08±11.34		37.71±11.24		<.001
Psychiatric symptoms (M±SD score)b
Affect	.26±.89		.31±.90		.225
Positive symptoms	.09±.93		.18±.94		.116
Negative symptoms	−.07±.92		.02±.95		<.001
Disorganized cognitive processing	−.02±.88		.07±.91		<.001
Violence factor score (violence in past 6 months) (M±SD)c	−.28±.71		−.29±.70		.493
Victimization factor score (victimization in past 6 months) (M±SD)d	−.19±.82		−.18±.82		.356

^aMeans were compared by t tests, and proportions were compared by chi square tests.

^bScores ranged from −1.50 to 3.20 for affect, −1.35 to 3.33 for positive symptoms, −1.72 to 3.36 for negative symptoms, and −1.63 to 3.67 for disorganized cognitive processing, with higher scores indicating greater symptomatology.

^cScores ranged from −.64 to 2.96, with higher scores indicating greater prevalence or severity of violence.

^dScores ranged from −.70 to 2.62, with higher scores indicating greater prevalence or severity of violent victimization.

Bivariate Analyses

The mean, standard deviation, and range of six-month violence factor scores, overall and across participant characteristics, are presented in Table 2. Results of bivariate analyses showed that all predictor variables were associated with short-term violence, with the exception of race-ethnicity. Age was negatively correlated with violence. Compared with female participants at follow-up, male participants reported less violence. Participants with a primary diagnosis of schizophrenia reported less violence than participants with other diagnoses. Participants with bipolar disorder reported less violence than those with major depression and those with a substance use disorder. Nonproblematic and problematic use of both alcohol and drugs were positively associated with violence. Affect was positively correlated with violence, whereas all other psychiatric symptoms exhibited negative correlations with the outcome variable. Recent hospitalization and baseline violence and victimization were positively correlated with violence.

TABLE 2.

Six-month violence risk factor scores in a follow-up sample of 2,579 participants with mental illnesses

Characteristic	Violence factor scorea			Test statistic	df	p
	M	SD	Range
Overall score for follow-up sample	−.39	.57	−.64 to 3.38
Sex				F=15.49	1, 2,577	<.001
Male	−.42	.53	−.64 to 3.38
Female	−.33	.63	−.64 to 2.64
Race-ethnicity				F=2.02	3, 2,575	.109
White	−.38	.54	−.64 to 2.79
Black	−.38	.61	−.64 to 3.38
Hispanic	−.44	.54	−.64 to 1.98
Other	−.55	.38	−.64 to 1.25
Primary diagnosis				F=62.40	4, 2,567	<.001
Schizophrenia	−.50	.47	−.64 to 3.38
Bipolar disorder	−.33	.61	−.64 to 2.64
Major depression	−.15	.68	−.64 to 2.79
Substance use disorder	−.05	.68	−.64 to 2.62
Other	−.16	.57	−.64 to 1.39
Alcohol use				F=40.21	2, 2,570	<.001
Abstinence	−.47	.49	−.64 to 2.62
Nonproblematic use	−.37	.56	−.64 to 2.62
Problematic use	−.24	.68	−.64 to 3.38
Drug use				F=28.83	2, 2,566	<.001
Abstinence	−.45	.52	−.64 to 2.64
Nonproblematic use	−.35	.57	−.64 to 2.24
Problematic use	−.25	.68	−.64 to 3.38
Recent hospitalization (past 6 months)				F=114.76	1, 2,576	<.001
No	−.51	.49	−.64 to 3.38
Yes	−.27	.62	−64 to 2.79
Age				r=−.22		<.001
Psychiatric symptoms
Affect				r=.15		<.001
Positive symptoms				r=−.08		<.001
Negative symptoms				r=−.14		<.001
Disorganized cognitive processing				r=−.15		<.001
Recent violence (past 6 months)				r=.40		<.001
Recent victimization (past 6 months)				r=.32		<.001

^aHigher violence factor scores indicate greater prevalence or severity of violence.

Multivariable Analyses

Diagnostic tests showed that tolerance values were all .420 or higher and variation inflation factor values were all below 2.379, indicating no multicollinearity among predictor variables (36). Model fit statistics and incremental validity of predictors across steps are presented in Table 3. Overall, results show improved prediction with each step, indicating that proximal indicators contributed to the prediction of short-term community violence above and beyond the static characteristics included in the regression model.

TABLE 3.

Hierarchical linear regression of predictors of community violence in the six months after baseline among 2,579 participants with mental illnesses

Model step	B	SE	t	p
Step 1. patient characteristicsa
Age	−.01	.00	−6.30	<.001
Male (reference: female)	−.06	.02	−2.67	.008
Race-ethnicity (reference: white)
Black	.07	.02	2.76	.006
Hispanic	.06	.05	1.14	.255
Other	−.02	.08	−.20	.843
Primary diagnosis (reference: schizophrenia)
Bipolar disorder	.10	.05	2.26	.024
Major depression	.26	.04	7.20	<.001
Substance use disorder	.36	.05	8.03	<.001
Other	.21	.12	1.69	.092
Step 2. substance useb
Alcohol use	.04	.02	3.02	.003
Drug use	.04	.02	2.42	.016
Step 3. clinical functioningc
Psychiatric symptoms
Affect	.03	.01	2.17	.030
Positive symptoms	.05	.02	3.20	<.001
Negative symptoms	.00	.02	−.21	.832
Disorganized cognitive processing	−.01	.02	−.59	.479
Recent hospitalization (reference: no)	.01	.03	.17	.868
Step 4. recent violence (in 6 months prior to baseline) (reference: no)d	.26	.02	16.42	<.001
Step 5. recent victimization (in 6 months prior to baseline) (reference: no)e	.05	.02	3.00	.003

^aF=31.26, df=10 and 2532, p<.001; adjusted R²=.106

^bF=28.21, df=12 and 2530, p<.001; adjusted R²=.114; ΔF=11.62, df=2 and 2530, p<.001; Δadjusted R²=.008

^cF=23.16, df=17 and 2525, p<.001; adjusted R²=.119; ΔF=3.97, df=5 and 2525, p<.001; Δadjusted R²=.005

^dF=38.65, df=18 and 2524, p<.001; adjusted R²=.204; ΔF=269.74, df=1 and 2524, p<.001; Δadjusted R²=.085

^eF=37.16, df=19 and 2523, p<.001; adjusted R²=.206; ΔF=9.01, df=1 and 2523, p=.003; Δadjusted R²=.002

Table 4 shows the final step of the model. In the final step, young age, female sex, and primary diagnoses of major depression and substance use disorder (compared with schizophrenia) were positive predictors of violence. Alcohol use, but not drug use, was a positive predictor of violence. None of the psychiatric symptoms were significant predictors. Baseline violence and victimization each significantly predicted six-month violence.

TABLE 4.

Full linear regression model of predictors of community violence in the six months after baseline among 2,579 participants with mental illnesses^a

Variable	B	SE	t	p
Age	.00	.00	−3.35	<.001
Male (reference: female)	−.05	.02	−2.41	.016
Race-ethnicity (reference: white)
Black	.04	.02	1.63	.104
Hispanic	.06	.05	1.16	.245
Other	−.02	.07	−.33	.742
Primary diagnosis (reference: schizophrenia)
Bipolar disorder	.08	.04	1.81	.070
Major depression	.19	.04	4.57	<.001
Substance use disorder	.24	.05	4.77	<.001
Other	.10	.12	.84	.399
Alcohol use (reference: abstinence)	.03	.01	2.17	.030
Drug use (reference: abstinence)	.01	.01	.35	.723
Psychiatric symptoms
Affect	.01	.01	1.09	.278
Positive symptoms	.03	.01	1.92	.055
Negative symptoms	.01	.01	.47	.638
Disorganized cognitive processing	.00	.02	−.14	.885
Recent hospitalization (reference: no)	−.04	.03	−1.47	.142
Committed recent violence (reference: no)	.23	.02	12.72	<.001
Experienced recent victimization (reference: no)	.05	.02	3.00	.003

^aF=37.16, df=19 and 2523, p<.001; adjusted R²=.206

DISCUSSION

Clinical assessments of violence risk inform delivery of mental health services, including treatment approaches, risk management strategies, and mandated treatment orders. This study examined individual and combined effects of static characteristics and of proximal, clinically relevant indicators of violence risk in a large, heterogeneous sample of adults with mental illnesses. Consistent with prior research, we found empirical support for the role of static characteristics, including age, sex, and primary diagnosis, in predicting community violence; however, alcohol use, violence, and victimization predicted subsequent violence, even after the analysis controlled for static characteristics. These findings add to the empirical evidence supporting the role of proximal factors in the assessment and management of short-term violence risk among adults with mental illnesses (15,37,38). Below, we discuss the observed effects of these proximal factors and how results may inform the administration of risk assessment instruments and formulation of interventions.

Although there is consensus on the increased risk of violence associated with substance use among adults with mental illnesses, extant findings are mixed regarding which indicator—alcohol use or drug use—is the more robust predictor (15,39,40). In this study’s multivariable models, alcohol use emerged as the better predictor of short-term community violence among adults with mental illnesses. (However, the bivariate relationship between drug use and subsequent violence was significant.) As such, integrated interventions targeting mental illness and alcohol use should reduce community-based violence risk in this population. These distinctions between alcohol and drug use become important when clinical efforts are not just broadly focused on improving psychosocial functioning among psychiatric patients but are also focused on reducing dangerousness to others (5,6).

Bivariate findings regarding the associations between clinical functioning and short-term violence risk underscore the importance of precise measurement and specification of psychiatric symptoms in research and practice. Although psychiatric symptoms failed to maintain significance in the multivariable model, their individual relationships with short-term violence remain relevant to clinical practice. Prior research on the associations between affect, positive symptoms, negative symptoms, and disorganized cognitive processing symptoms and violence perpetration has sometimes been inconsistent (41,42). This may result, in part, from examination of these symptoms as individual predictors, as well as from differences in assessment and follow-up time frames (16). Indeed, most significant associations are found in cross-sectional structures, which do not ensure temporality (Van Dorn et al., unpublished manuscript, 2015). The findings of this study suggest that consideration of current psychiatric symptoms, instead of psychiatric diagnosis alone, should assist in the identification of individuals at heightened risk of short-term violence. Furthermore, evidence-based treatment targeting psychiatric symptoms, including cognitive, behavioral, and psychopharmacological interventions, should contribute to reductions in violence risk (43–46). Beyond these main effects, however, there remains a need for research examining how these symptoms may interact with one another to increase violence risk (47).

The strong effect of past violence provides further evidence that a history of violence is a key risk factor for future violence but also suggests that recent violence, specifically, should be attended to in clinical assessments of short-term risk. The same is true for victimization; in fact, recent victimization, although rarely examined in research, added to the prediction of short-term violence when the analysis controlled for all other factors in the model, including recent violence. This finding, consistent with that of Sadeh and colleagues (24) regarding the role of victimization in predicting long-term (that is, 12-month) violence, provides compelling evidence that recent victimization should be considered in the assessment of both short- and long-term violence risk. However, to our knowledge, only one instrument, the Short-Term Assessment of Risk and Treatability (START) (9), considers both recent violence and recent victimization. Moreover, there is much discussion of and emphasis on trauma-informed care in this population, and among female psychiatric patients in particular, although the focus is typically on childhood victimization experiences (48,49). Findings suggest that trauma-informed approaches should also consider the trauma and sequelae of adult victimization experiences and that these efforts may reduce risk of violence.

This study had several strengths, including its large, representative sample; inclusion of proximal, clinically relevant indicators; and prospective data structure. However, findings should be interpreted within the context of the study’s limitations. First, our data on violence and victimization were derived from self-report and may thus be susceptible to the effects of social desirability, recall bias, and errors. Although self-report is a valid and reliable approach for collecting data on violence and victimization (50,51), additional sources of information across all parent studies, such as hospital and arrest records, may have assisted in capturing non–self-reported violent events. Second, attrition across studies over time resulted in missing outcome data for a large portion of the integrated sample (N=1,905, 55%). Nevertheless, this study remains the largest examination of the effects of proximal indicators on short-term community violence in this population. Third, the assessment periods implemented across parent studies limited our evaluation of even shorter-term effects. Specifically, we operationalized recent, current, and short-term variables as occurring within a six-month time frame; however, the optimal assessment and prediction time frame for these variables is unknown.

CONCLUSIONS

This study provides empirical evidence for three proximal, clinically relevant indicators in the assessment and management of short-term violence risk among adults with mental illnesses: current alcohol use, recent violence, and recent victimization. Although clinicians may never be able to answer the public calls for the absolute prediction—and prevention—of violence by adults with mental illnesses (6), attending to these indicators in clinical practice should assist in the identification of persons at heightened risk of community-based violence. There is now a preponderance of evidence demonstrating the superiority of structured over unstructured approaches to assessing violence risk (52,53). Clinicians should use a validated risk assessment instrument that emphasizes proximal indicators, such as the START or the HCR-20 (12), to reduce violence risk in this population (54–56).