10.1192/bjp.176.4.

312 Access the most recent version at DOI:

2000, 176:312-319. BJP
APPELBAUM, THOMAS GRISSO, EDWARD P. MULVEY and LOREN H. ROTH
JOHN MONAHAN, HENRY J. STEADMAN, PAMELA C. ROBBINS, ERIC SILVER, PAUL S.
violence risk
Developing a clinically useful actuarial tool for assessing
References
http://bjp.rcpsych.org/content/176/4/312#BIBL
This article cites 0 articles, 0 of which you can access for free at:
permissions
Reprints/
permissions@rcpsych.ac.uk write to
To obtain reprints or permission to reproduce material from this paper, please
to this article at
You can respond
http://bjp.rcpsych.org/letters/submit/bjprcpsych;176/4/312
from
Downloaded
The Royal College of Psychiatrists Published by
on July 15, 2014 http://bjp.rcpsych.org/
http://bjp.rcpsych.org/site/subscriptions/
go to: The British Journal of Psychiatry To subscribe to
Background Background Anewactuarial method Anewactuarial method
for violenceriskassessment ^ the for violenceriskassessment ^ the
Iterative ClassificationTree (ICT) ^ has Iterative ClassificationTree (ICT) ^ has
become available. It has a high degree of become available. It has a high degree of
accuracybut canbetime andresource accuracybut canbetime andresource
intensiveto administer. intensiveto administer.
Aims Aims Toincreasetheclinical utilityofthe Toincreasetheclinical utilityofthe
ICT methodbyrestricting therisk factors ICT methodbyrestricting therisk factors
usedtogeneratetheactuarial tool tothose usedtogeneratetheactuarial tool tothose
commonlyavailableinhospital records or commonlyavailableinhospital records or
capable of beingroutinelyassessedin capable of beingroutinelyassessedin
clinical practice. clinical practice.
Method Method Atotal of 939 male and female Atotal of 939 male and female
civil psychiatric patients between18 and civil psychiatric patients between18 and
40 years oldwere assessed on106 risk 40 years oldwere assessed on106 risk
factorsinthe hospital andmonitored for factors inthe hospital andmonitored for
violenceto others during the first 20 violenceto others during the first 20
weeks after discharge. weeks afterdischarge.
Results Results The ICT classified 72.6% of the The ICT classified 72.6% of the
sample as eitherlowrisk (less thanhalf of sample as eitherlowrisk (less thanhalf of
the sample's baserate of violence) or high the sample's baserate of violence) or high
risk (morethantwicethe sample's base risk (morethantwicethe sample's base
rate of violence). rate of violence).
Conclusions Conclusions Aclinicallyuseful Aclinicallyuseful
actuarial methodexists toassistinviolence actuarial methodexists toassistinviolence
riskassessment. riskassessment.
Declaration of interest Declaration of interest Research Research
was fundedby the John D. and Catherine was fundedby the John D. and Catherine
T. MacArthur Foundation andby the US T. MacArthur Foundation andby the US
National Institute of Mental Health. National Institute of Mental Health.
A recent survey in the USA found that 95% A recent survey in the USA found that 95%
of the general public believes that when a of the general public believes that when a
person with a mental disorder is predicted person with a mental disorder is predicted
to be violent to others, legal intervention to be violent to others, legal intervention
to avert the anticipated harm is justified to avert the anticipated harm is justified
(Pescosolido (Pescosolido et al et al, 1999). The pervasiveness , 1999). The pervasiveness
of such belief throughout the world helps to of such belief throughout the world helps to
explain the wide and growing variety of explain the wide and growing variety of
laws, including in-patient and out-patient laws, including in-patient and out-patient
civil commitment, sexual predator commit- civil commitment, sexual predator commit-
ment, tort liability and employment discri- ment, tort liability and employment discri-
mination, that call upon psychiatrists and mination, that call upon psychiatrists and
psychologists to assess the risk of violence psychologists to assess the risk of violence
(Reed, 1997). A large body of research con- (Reed, 1997). A large body of research con-
ducted in the 1970s and 1980s called into ducted in the 1970s and 1980s called into
question the ability of clinicians to make question the ability of clinicians to make
accurate risk assessments of violence of accurate risk assessments of violence of
the type that the public and the law de- the type that the public and the law de-
mand. More recent research (Lidz mand. More recent research (Lidz et al et al, ,
1993) has been only slightly more sanguine. 1993) has been only slightly more sanguine.
One repeatedly suggested way to improve One repeatedly suggested way to improve
the accuracy of risk assessments of violence the accuracy of risk assessments of violence
has been to use statistical or actuarial meth- has been to use statistical or actuarial meth-
ods to inform clinical judgement (Borum, ods to inform clinical judgement (Borum,
1996). However, where clinically relevant 1996). However, where clinically relevant
actuarial tools have been available, their actuarial tools have been available, their
application has been sufficiently cumber- application has been sufficiently cumber-
some and time-consuming that actuarial in- some and time-consuming that actuarial in-
put into the risk assessment of violence has put into the risk assessment of violence has
been impractical in most real-world clinical been impractical in most real-world clinical
settings (Gardner settings (Gardner et al et al, 1996). , 1996).
ITERATIVE CLASSIFICATION ITERATIVE CLASSIFICATION
TREE METHOD TREE METHOD
We have recently developed an actuarial We have recently developed an actuarial
tool for assessing the risk of violence by tool for assessing the risk of violence by
people discharged from psychiatric facilities people discharged from psychiatric facilities
that we believe has greater potential for that we believe has greater potential for
clinical application than existing actuarial clinical application than existing actuarial
methods (Steadman methods (Steadman et al et al, 2000). We call , 2000). We call
this tool the Iterative Classification Tree this tool the Iterative Classification Tree
(ICT). Classification trees (Breiman (ICT). Classification trees (Breiman et al et al, ,
1984) have long been used in a number of 1984) have long been used in a number of
areas of medicine, including neurology areas of medicine, including neurology
(e.g. Levy (e.g. Levy et al et al, 1985). A classification tree , 1985). A classification tree
approach to the risk assessment of violence approach to the risk assessment of violence
is predicated upon an interactive and con- is predicated upon an interactive and con-
tingent model of violence: one that allows tingent model of violence: one that allows
many different combinations of risk factors many different combinations of risk factors
to classify a person as high or low risk. to classify a person as high or low risk.
Whether a particular question is asked in Whether a particular question is asked in
any clinical assessment grounded in this ap- any clinical assessment grounded in this ap-
proach depends on the answers given to proach depends on the answers given to
each prior question. Based on a sequence each prior question. Based on a sequence
established by the classification tree, a first established by the classification tree, a first
question is asked of all persons being as- question is asked of all persons being as-
sessed. Contingent on the answer to that sessed. Contingent on the answer to that
question, one or another second question question, one or another second question
is posed, and so on, until each person is is posed, and so on, until each person is
classified into a category on the basis of classified into a category on the basis of
the risk of violence. This contrasts with the risk of violence. This contrasts with
the usual approach to actuarial risk assess- the usual approach to actuarial risk assess-
ment in which a common set of questions is ment in which a common set of questions is
asked of everyone being assessed and every asked of everyone being assessed and every
answer is weighted and summed to produce answer is weighted and summed to produce
a score that can be used for the purposes of a score that can be used for the purposes of
categorisation. categorisation.
In addition to its tree-based character, In addition to its tree-based character,
our approach acknowledges the practical our approach acknowledges the practical
impossibility of adequately classifying all impossibility of adequately classifying all
persons into a high or a low violence risk persons into a high or a low violence risk
group. Therefore, rather than relying on group. Therefore, rather than relying on
the standard single threshold for distin- the standard single threshold for distin-
guishing among cases, our approach em- guishing among cases, our approach em-
ploys two thresholds: one for identifying ploys two thresholds: one for identifying
high-risk cases and one for identifying high-risk cases and one for identifying
low-risk cases. We assume that inevitably low-risk cases. We assume that inevitably
there will be cases that fall between these there will be cases that fall between these
two thresholds, cases for which any predic- two thresholds, cases for which any predic-
tion scheme is incapable of making an ade- tion scheme is incapable of making an ade-
quate assessment of high or low risk. Based quate assessment of high or low risk. Based
on current knowledge, the aggregate degree on current knowledge, the aggregate degree
of risk presented by these intermediate of risk presented by these intermediate
cases cannot be distinguished statistically cases cannot be distinguished statistically
from the base rate of the sample as a whole. from the base rate of the sample as a whole.
CLINICALUTILITY CLINICALUTILITY
Our first test of the ICT method (Steadman Our first test of the ICT method (Steadman
et al et al, 2000) focused on how well the meth- , 2000) focused on how well the meth-
od performed in making risk assessments of od performed in making risk assessments of
violence under ideal conditions (i.e. with few violence under ideal conditions (i.e. with few
constraints on the time or resources neces- constraints on the time or resources neces-
sary to gather risk factors). For example, sary to gather risk factors). For example,
the risk factor that most clearly differen- the risk factor that most clearly differen-
tiated high-risk from low-risk groups was tiated high-risk from low-risk groups was
the Hare Psychopathy Checklist: Screening the Hare Psychopathy Checklist: Screening
Version (Hare PCLSV; Hart Version (Hare PCLSV; Hart et al et al, 1995). , 1995).
Given that the full Hare PCLR requires Given that the full Hare PCLR requires
several hours to administer the screening several hours to administer the screening
version alone takes over one hour to admin- version alone takes over one hour to admin-
ister and that it has to be administered by ister and that it has to be administered by
experienced clinicians whom Hare (1998) experienced clinicians whom Hare (1998)
recommended should undergo three days recommended should undergo three days
of specialised training, resource constraints of specialised training, resource constraints
312 312
BRI TI S H J OURNAL OF P SYCHI ATRY BRI TI S H J OURNAL OF P SYCHI ATRY ( 2 0 0 0 ) , 17 6 , 3 1 2 ^ 3 1 9 ( 2 0 0 0 ) , 1 7 6 , 3 1 2 ^ 31 9
Developing a clinically useful actuarial tool Developing a clinically useful actuarial tool
for assessing violence risk for assessing violence risk
y
JOHN MONAHAN, HENRY J. STEADMAN, PAUL S. APPELBAUM, JOHN MONAHAN, HENRY J. STEADMAN, PAUL S. APPELBAUM,
PAMELA C. ROBBINS, EDWARD P. MULVEY, ERIC SILVER, LOREN H. ROTH PAMELA C. ROBBINS, EDWARD P. MULVEY, ERIC SILVER, LOREN H. ROTH
and THOMAS GRISSO and THOMAS GRISSO
y
See editorial pp. 307^311, this issue See editorial pp. 307^311, this issue
ACTUARI AL TOOL FOR AS S ES S ING VIOLENCE RI SK ACTUARI AL TOOL F OR AS S ES SING VIOLENCE RI S K
in many non-forensic clinical settings will in many non-forensic clinical settings will
preclude its use. preclude its use.
This article applies the ICT method to This article applies the ICT method to
the sample of patients assessed in the the sample of patients assessed in the
MacArthur Violence Risk Assessment MacArthur Violence Risk Assessment
Study (Steadman Study (Steadman et al et al, 1998). Our goal is , 1998). Our goal is
to increase the clinical utility of this actuar- to increase the clinical utility of this actuar-
ial method by restricting the risk factors ial method by restricting the risk factors
tested to those commonly available in hos- tested to those commonly available in hos-
pital records or capable of being assessed pital records or capable of being assessed
routinely in clinical practice. routinely in clinical practice.
METHOD METHOD
Subject enrolment Subject enrolment
Admissions were sampled from acute psy- Admissions were sampled from acute psy-
chiatric in-patient facilities at three sites: chiatric in-patient facilities at three sites:
Western Psychiatric Institute and Clinic Western Psychiatric Institute and Clinic
(Pittsburgh, PA); Western Missouri Mental (Pittsburgh, PA); Western Missouri Mental
Health Center (Kansas City, MO); Wor- Health Center (Kansas City, MO); Wor-
cester State Hospital and the University of cester State Hospital and the University of
Massachusetts Medical Center (Worcester, Massachusetts Medical Center (Worcester,
MA). Selection criteria for research subjects MA). Selection criteria for research subjects
were: civil admissions; between the ages of were: civil admissions; between the ages of
18 and 40 years; English-speaking; White 18 and 40 years; English-speaking; White
or African American ethnicity (or Hispanic or African American ethnicity (or Hispanic
in Worcester only); and a chart diagnosis of in Worcester only); and a chart diagnosis of
schizophrenia, schizophreniform disorder, schizophrenia, schizophreniform disorder,
schizoaffective disorder, depression, dys- schizoaffective disorder, depression, dys-
thymia, mania, brief reactive psychosis, thymia, mania, brief reactive psychosis,
delusional disorder, alcohol or drug abuse delusional disorder, alcohol or drug abuse
or dependence, or a personality disorder. or dependence, or a personality disorder.
After complete description of the study to After complete description of the study to
the subjects, written informed consent was the subjects, written informed consent was
obtained. obtained.
Sample description Sample description
We approached a quota sample (to ensure We approached a quota sample (to ensure
representativeness across sites on gender, representativeness across sites on gender,
race, and age) of 1695 to participate. The race, and age) of 1695 to participate. The
refusal rate was 29% ( refusal rate was 29% (n n492). The final 492). The final
sample given a hospital interview was sample given a hospital interview was
1136. Differences between the eligible 1136. Differences between the eligible
admissions and the follow-up sample admissions and the follow-up sample
( (n n939) are discussed in detail elsewhere 939) are discussed in detail elsewhere
(Steadman (Steadman et al et al, 1998). Males comprised , 1998). Males comprised
57.3% of the sample. Ethnically, 68.7% 57.3% of the sample. Ethnically, 68.7%
of the sample was White, 29.1% African of the sample was White, 29.1% African
American and 2.2% Hispanic. The mean American and 2.2% Hispanic. The mean
age was 29.9 (s.d. age was 29.9 (s.d.6.2) years. Depression 6.2) years. Depression
was the most frequent primary research was the most frequent primary research
diagnosis on the DSMIIIR Checklist diagnosis on the DSMIIIR Checklist
(Janca & Helzer, 1990; 41.9%), followed (Janca & Helzer, 1990; 41.9%), followed
by alcohol/drug abuse or dependence by alcohol/drug abuse or dependence
(21.8%), schizophrenia (17.0%), bipolar (21.8%), schizophrenia (17.0%), bipolar
disorder (14.1%), personality disorder only disorder (14.1%), personality disorder only
(2.1%) and other psychotic disorder (2.1%) and other psychotic disorder
(3.1%). The proportion of all cases with a (3.1%). The proportion of all cases with a
primary research diagnosis of major mental primary research diagnosis of major mental
313 313
Table 1 Table 1 Risk factors and bivariate correlations with violence in the first two follow-ups Risk factors and bivariate correlations with violence in the first two follow-ups
Domain Domain Reference Reference Pearson coefficient Pearson coefficient R R
Personal Personal
Gender: male Gender: male 0.08* 0.08*
Age Age 7 70.07* 0.07*
Race: White Race: White 7 70.12*** 0.12***
Verbal IQ Verbal IQ
1 1
7 70.11** 0.11**
Ever married Ever married 0.01 0.01
Hare Psychopathy Checklist: Hare Psychopathy Checklist:
Screening Version Screening Version 4 412 12
1 1
Hart Hart et al et al, 1995 , 1995 0.26*** 0.26***
Novaco anger: behaviour Novaco anger: behaviour Novaco, 1994 Novaco, 1994 0.16*** 0.16***
Novaco anger: cognitive Novaco anger: cognitive 0.11* 0.11*
Novaco anger: arousal Novaco anger: arousal 0.09** 0.09**
Novaco anger: intensity Novaco anger: intensity 0.08* 0.08*
Barratt impulsiveness: motor Barratt impulsiveness: motor Barratt, 1994 Barratt, 1994 0.07* 0.07*
Barratt impulsiveness: non-planning Barratt impulsiveness: non-planning 0.05 0.05
Barratt impulsiveness: cognitive Barratt impulsiveness: cognitive 0.05 0.05
Historical Historical
Years of education Years of education 7 70.11*** 0.11***
Socio-economic status Socio-economic status
1 1
Hollingshead & Redlich, 1958 Hollingshead & Redlich, 1958 0.05 0.05
Employed Employed 7 70.05 0.05
Age at first hospitalisation Age at first hospitalisation 7 70.04 0.04
No. of prior hospitalisations No. of prior hospitalisations 7 70.03 0.03
Involuntary legal status Involuntary legal status 0.11** 0.11**
Recent violent behaviour Recent violent behaviour 0.14*** 0.14***
Adult arrest: seriousness Adult arrest: seriousness 0.25*** 0.25***
Adult arrest: frequency Adult arrest: frequency 0.24*** 0.24***
Any arrest: person crime Any arrest: person crime
1 1
Official Report Official Report 0.13*** 0.13***
Any arrest: other crime Any arrest: other crime
1 1
Official Report Official Report 0.11*** 0.11***
Sexually abused before age 20 Sexually abused before age 20 7 70.03 0.03
Seriousness of abuse as child Seriousness of abuse as child 0.14*** 0.14***
Frequency of abuse as child Frequency of abuse as child 0.12*** 0.12***
Father ever used drugs Father ever used drugs 0.16*** 0.16***
Father ever arrested Father ever arrested 0.15*** 0.15***
Father ever excess drinking Father ever excess drinking 0.11** 0.11**
Father ever admitted to psychiatric hospital Father ever admitted to psychiatric hospital 0.02 0.02
Lived with father to age 15 Lived with father to age 15 7 70.09** 0.09**
Mother ever used drugs Mother ever used drugs 0.05 0.05
Mother ever arrested Mother ever arrested 0.05 0.05
Mother ever excess drinking Mother ever excess drinking 0.06 0.06
Mother ever admitted to psychiatric hospital Mother ever admitted to psychiatric hospital 7 70.02 0.02
Lived with mother to age 15 Lived with mother to age 15 7 70.06 0.06
Parents ever fought with each other Parents ever fought with each other 0.06 0.06
Parents ever fought with others Parents ever fought with others 0.03 0.03
Any head injury: loss of consciousness Any head injury: loss of consciousness 0.10** 0.10**
Any head injury: no loss of consciousness Any head injury: no loss of consciousness 0.06 0.06
Self-harm thoughts Self-harm thoughts 0.02 0.02
Self-harm attempt Self-harm attempt 7 70.03 0.03
Attempt to kill self Attempt to kill self 0.01 0.01
Contextual Contextual
Living in private residence Living in private residence 7 70.05 0.05
Homeless Homeless 0.05 0.05
Living alone Living alone 7 70.07* 0.07*
(continued) (continued)
MONAHAN E T AL MONAHAN E T AL
disorder that had a co-occurring diagnosis disorder that had a co-occurring diagnosis
of substance abuse or dependence was as of substance abuse or dependence was as
follows: depression, 49.6%; schizophrenia, follows: depression, 49.6%; schizophrenia,
41%; 41%; bipolar disorder, 37.7%; and other bipolar disorder, 37.7%; and other
psychotic psychotic disorder, 45%. disorder, 45%.
Hospital data collection Hospital data collection
Hospital data collection was conducted in Hospital data collection was conducted in
two parts: an interview by a research inter- two parts: an interview by a research inter-
viewer to obtain data on risk factors and viewer to obtain data on risk factors and
violence; and an interview by a research violence; and an interview by a research
clinician (PhD or MA/MSW in psychology clinician (PhD or MA/MSW in psychology
or social work) to confirm the chart diag- or social work) to confirm the chart diag-
nosis using the DSMIIIR Checklist and nosis using the DSMIIIR Checklist and
to administer several clinical instruments. to administer several clinical instruments.
The hospital data set assembled in the The hospital data set assembled in the
MacArthur Violence Risk Assessment MacArthur Violence Risk Assessment
Study consisted of 134 risk factors from Study consisted of 134 risk factors from
four conceptual domains: dispositional or four conceptual domains: dispositional or
personal factors (e.g. age); historical or de- personal factors (e.g. age); historical or de-
velopmental factors (e.g. child abuse); con- velopmental factors (e.g. child abuse); con-
textual or situational factors (e.g. social textual or situational factors (e.g. social
networks); and clinical or symptom factors networks); and clinical or symptom factors
(e.g. delusions) (Steadman (e.g. delusions) (Steadman et al et al, 1994). For , 1994). For
the present analysis, we eliminated 28 risk the present analysis, we eliminated 28 risk
factors that would be the most difficult to factors that would be the most difficult to
obtain in clinical practice, restricting our- obtain in clinical practice, restricting our-
selves to the remaining 106. Two criteria selves to the remaining 106. Two criteria
were used to eliminate risk factors. The first were used to eliminate risk factors. The first
was to eliminate information generally un- was to eliminate information generally un-
available to mental health personnel in the available to mental health personnel in the
context of brief hospitalisation (e.g. infor- context of brief hospitalisation (e.g. infor-
mation in official arrest records, in distinc- mation in official arrest records, in distinc-
tion to self-report of prior arrests). The tion to self-report of prior arrests). The
second was to eliminate information that second was to eliminate information that
required the administration of a lengthy required the administration of a lengthy
( (4 412-item) instrument to obtain (e.g. a 12-item) instrument to obtain (e.g. a
social network inventory (Estroff & Zim- social network inventory (Estroff & Zim-
mer, 1994)). A list of all 134 risk factors, mer, 1994)). A list of all 134 risk factors,
with their bivariate correlations with vio- with their bivariate correlations with vio-
lence and with an indication of which were lence and with an indication of which were
eliminated from these analyses, is provided eliminated from these analyses, is provided
in Table 1. in Table 1.
Community data collection Community data collection
Twenty weeks after hospital discharge was Twenty weeks after hospital discharge was
chosen as the time frame for the analysis chosen as the time frame for the analysis
here because this was the period during here because this was the period during
which the prevalence of violence by pa- which the prevalence of violence by pa-
tients in the community was at its highest tients in the community was at its highest
(Steadman (Steadman et al et al, 1998). Research inter- , 1998). Research inter-
viewers attempted two follow-up inter- viewers attempted two follow-up inter-
views with enrolled patients in the views with enrolled patients in the
community during this period, approxi- community during this period, approxi-
mately 10 weeks apart. A collateral infor- mately 10 weeks apart. A collateral infor-
mant who knew of the patient's behaviour mant who knew of the patient's behaviour
in the community during the follow-up in the community during the follow-up
period usually, but not always, a family period usually, but not always, a family
member was also interviewed on the member was also interviewed on the
314 314
Table 1 Table 1 (continued) (continued)
Domain Domain Reference Reference Pearson coefficient Pearson coefficient R R
Perceived stress Perceived stress
1 1
Cohen Cohen et al et al, 1983 , 1983 0.08* 0.08*
Social networks Social networks Estroff & Zimmer, 1994 Estroff & Zimmer, 1994
No. of people in social network No. of people in social network
1 1
7 70.02 0.02
% Mental health professionals in social network % Mental health professionals in social network
1 1
7 70.10** 0.10**
% Family in social network % Family in social network
1 1
0.01 0.01
No. of negative persons in social network No. of negative persons in social network
1 1
0.07* 0.07*
No. of positive and material supporters No. of positive and material supporters
1 1
7 70.07* 0.07*
Average no. of mentions per negative person Average no. of mentions per negative person
1 1
0.06 0.06
Average no. of mentions per positive material Average no. of mentions per positive material
person person
1 1
7 70.03 0.03
Frequency of social network contact Frequency of social network contact
1 1
7 70.03 0.03
Duration of social network contact Duration of social network contact
1 1
0.02 0.02
Clinical Clinical
Chart antisocial personality disorder Chart antisocial personality disorder 0.19*** 0.19***
DSM^III^R Checklist DSM^III^R Checklist Janca & Helzer, 1990 Janca & Helzer, 1990
Major disorder, no substance Major disorder, no substance 7 70.19*** 0.19***
Major disorder and substance Major disorder and substance 0.08* 0.08*
Substance, no major disorder Substance, no major disorder 0.15*** 0.15***
Drug or alcohol Drug or alcohol 0.18*** 0.18***
Drug Drug 0.17*** 0.17***
Alcohol Alcohol 0.14*** 0.14***
Schizophrenia Schizophrenia 7 70.12*** 0.12***
Mania Mania 7 70.04 0.04
Depression Depression 7 70.02 0.02
Other psychosis Other psychosis 0.00 0.00
Personality disorder only Personality disorder only 0.02 0.02
Brief Psychiatric Rating Scale Brief Psychiatric Rating Scale Overall, 1988 Overall, 1988
Total score Total score
1 1
7 70.04 0.04
Activation sub-scale Activation sub-scale
1 1
7 70.08* 0.08*
Hostility sub-scale Hostility sub-scale
1 1
0.08* 0.08*
Anergia sub-scale Anergia sub-scale
1 1
7 70.07* 0.07*
Thought disturbance sub-scale Thought disturbance sub-scale
1 1
7 70.06* 0.06*
Anxiety/depression sub-scale Anxiety/depression sub-scale
1 1
0.01 0.01
Global Assessment of Functioning Global Assessment of Functioning American Psychiatric American Psychiatric
Association, 1989 Association, 1989
7 70.05 0.05
Activities of daily living Activities of daily living 7 70.01 0.01
Delusions Delusions Appelbaum Appelbaum et al et al, 1999 , 1999
Any delusions Any delusions
1 1
7 70.06 0.06
Persecutory Persecutory 7 70.07* 0.07*
Grandiose Grandiose 7 70.01 0.01
Body/mind control Body/mind control 7 70.09** 0.09**
Thought broadcasting Thought broadcasting 7 70.05 0.05
Religious Religious 7 70.08* 0.08*
Jealousy Jealousy 7 70.02 0.02
Guilt Guilt 7 70.03 0.03
Somatic Somatic 7 70.03 0.03
Influence on others Influence on others 7 70.03 0.03
Threat/control-override Threat/control-override 7 70.10** 0.10**
Other Other 7 70.04 0.04
(continued) (continued)
ACTUARI AL TOOL FOR AS S ES S ING VIOLENCE RI SK ACTUARI AL TOOL F OR AS S ES SING VIOLENCE RI S K
same schedule. Arrest and re-hospitalisa- same schedule. Arrest and re-hospitalisa-
tion records provided the third source of in- tion records provided the third source of in-
formation about the patients' behaviour in formation about the patients' behaviour in
the community. the community.
Patients and collaterals independently Patients and collaterals independently
were asked whether the patient had been were asked whether the patient had been
involved in any of several categories of vio- involved in any of several categories of vio-
lent behaviour in the past 10 weeks (Lidz lent behaviour in the past 10 weeks (Lidz et et
al al, 1993). Only the most serious act for , 1993). Only the most serious act for
each discrete incident was coded. Violence each discrete incident was coded. Violence
to others was defined to include the follow- to others was defined to include the follow-
ing: acts of battery that resulted in physical ing: acts of battery that resulted in physical
injury; sexual assaults; assaultive acts that injury; sexual assaults; assaultive acts that
involved the use of a weapon; or threats involved the use of a weapon; or threats
made with a weapon in hand. (Battery that made with a weapon in hand. (Battery that
did not result in injury was defined as did not result in injury was defined as
`other aggressive act' (Steadman `other aggressive act' (Steadman et al et al, ,
1998) and is not considered in the analyses 1998) and is not considered in the analyses
reported here.) Violence reported by any of reported here.) Violence reported by any of
the three data sources subject self-report, the three data sources subject self-report,
collateral report, or official records was collateral report, or official records was
reviewed by a team of trained coders. Ethi- reviewed by a team of trained coders. Ethi-
cal and legal issues encountered in conduct- cal and legal issues encountered in conduct-
ing this research are discussed elsewhere ing this research are discussed elsewhere
(Monahan (Monahan et al et al, 1994). , 1994).
Developing the classification tree Developing the classification tree
To develop the ICT model, we used To develop the ICT model, we used
CHAID (chi-squared automatic interaction CHAID (chi-squared automatic interaction
detector) software (SPSS, 1993). Specifi- detector) software (SPSS, 1993). Specifi-
cally, the CHAID algorithm was used to cally, the CHAID algorithm was used to
assess the statistical significance of the assess the statistical significance of the
bivariate association between each of the bivariate association between each of the
106 eligible risk factors and the dichoto- 106 eligible risk factors and the dichoto-
mous outcome measure violence in the mous outcome measure violence in the
community until the most statistically community until the most statistically
significant value of significant value of w w
2 2
was identified, with was identified, with
P P5 50.05 a necessary condition for risk 0.05 a necessary condition for risk
factor selection. Once a risk factor was factor selection. Once a risk factor was
selected, the sample was partitioned selected, the sample was partitioned
according to the values of that risk factor. according to the values of that risk factor.
This selection procedure was then repeated This selection procedure was then repeated
for each of the sample partitions, thus for each of the sample partitions, thus
further partitioning the sample. The result further partitioning the sample. The result
of the partitioning process was to identify of the partitioning process was to identify
groups of cases that shared the same risk groups of cases that shared the same risk
factors and that also shared the same values factors and that also shared the same values
on the outcome measure of violence. on the outcome measure of violence.
Iterating the classification tree Iterating the classification tree
We then extended this recursive partition- We then extended this recursive partition-
ing approach in an iterative fashion. That ing approach in an iterative fashion. That
is, all subjects not classified into groups de- is, all subjects not classified into groups de-
signated as either high risk or low risk in signated as either high risk or low risk in
the first iteration of CHAID were pooled the first iteration of CHAID were pooled
together and re-analysed in a second itera- together and re-analysed in a second itera-
tion of CHAID. This iterative process con- tion of CHAID. This iterative process con-
tinued until it was not possible to classify tinued until it was not possible to classify
any additional groups of subjects as either any additional groups of subjects as either
315 315
Table 1 Table 1 (continued) (continued)
Domain Domain Reference Reference Pearson coefficient Pearson coefficient R R
Violent fantasies Violent fantasies Grisso Grisso et al et al, 2000 , 2000
Any Any 0.13*** 0.13***
Frequent Frequent 0.13*** 0.13***
Recent onset Recent onset 0.07* 0.07*
Same target Same target 0.03 0.03
Focus same person Focus same person 0.10** 0.10**
Escalating harm Escalating harm 0.13*** 0.13***
While with target While with target 0.12*** 0.12***
Frequent, not escalating, not with target Frequent, not escalating, not with target 7 70.01 0.01
Frequent, escalating, not with target Frequent, escalating, not with target 0.09** 0.09**
Frequent, not escalating, with target Frequent, not escalating, with target 0.08* 0.08*
Frequent, escalating, with target Frequent, escalating, with target 0.10** 0.10**
Not frequent, not escalating, not with target Not frequent, not escalating, not with target 0.13*** 0.13***
Any hallucinations Any hallucinations 0.02 0.02
Command hallucinations Command hallucinations 0.06 0.06
Present at time of admission Present at time of admission Record Review Record Review
Substance abuse Substance abuse 0.14*** 0.14***
Paranoia Paranoia 7 70.09** 0.09**
Delusions Delusions 7 70.09** 0.09**
Decompensation Decompensation 7 70.09** 0.09**
Violence Violence 0.09** 0.09**
Hallucinations Hallucinations 7 70.07* 0.07*
Bizarre behaviour Bizarre behaviour 7 70.07* 0.07*
Medication non-adherence Medication non-adherence 7 70.07* 0.07*
Aggressive (non-violent) Aggressive (non-violent) 0.06 0.06
Anxiety Anxiety 7 70.05 0.05
Suicide attempt Suicide attempt 0.05 0.05
Mania Mania 7 70.04 0.04
Personal problems Personal problems 0.03 0.03
Evaluation Evaluation 0.03 0.03
Other Other 7 70.03 0.03
Medication change Medication change 7 70.02 0.02
Unable to care for self Unable to care for self 0.02 0.02
Suicide threat Suicide threat 7 70.01 0.01
Property damage Property damage 7 70.01 0.01
Court order Court order 7 70.01 0.01
Depression Depression 7 70.003 0.003
Druguse Druguse
Any drug Any drug 0.12*** 0.12***
Cocaine Cocaine 0.11** 0.11**
Alcohol Alcohol 0.10** 0.10**
Other Other 0.08* 0.08*
Marijuana Marijuana 0.04 0.04
Stimulants Stimulants 0.04 0.04
Sedatives Sedatives 0.03 0.03
Opiates Opiates 0.04 0.04
Mini-Mental State Mini-Mental State
1 1
Folstein Folstein et al et al, 1975 , 1975 0.02 0.02
Perceived coercion at admission Perceived coercion at admission
1 1
Gardner Gardner et al et al, 1993 , 1993 0.03 0.03
Measures with no reference were obtained using project instruments and are available from the first author upon Measures with no reference were obtained using project instruments and are available from the first author upon
request. request.
* *P P5 50.05; ** 0.05; **P P5 50.01; *** 0.01; ***P P5 50.001. 0.001.
MONAHAN E T AL MONAHAN E T AL
high or low risk (with no group allowed to high or low risk (with no group allowed to
contain fewer than 50 cases). contain fewer than 50 cases).
Choosing two cut-offs Choosing two cut-offs
The choice of cut-off scores for high-risk The choice of cut-off scores for high-risk
and low-risk categories must be made in and low-risk categories must be made in
the context of legal or policy values exter- the context of legal or policy values exter-
nal to the methodology chosen for assessing nal to the methodology chosen for assessing
risk. Here, for illustrative purposes, we de- risk. Here, for illustrative purposes, we de-
fined any group of patients with a rate of fined any group of patients with a rate of
violence that was violence that was less than half less than half the base the base
prevalence rate of the total sample, as in prevalence rate of the total sample, as in
the low-risk category, and any group of pa- the low-risk category, and any group of pa-
tients whose rate of violence was tients whose rate of violence was greater greater
than twice than twice the base prevalence rate of the the base prevalence rate of the
total sample, as in the high-risk category. total sample, as in the high-risk category.
Because the base prevalence rate of violence Because the base prevalence rate of violence
during the first 20 weeks after hospital dis- during the first 20 weeks after hospital dis-
charge for the total sample was 18.7% (i.e. charge for the total sample was 18.7% (i.e.
18.7% of the patients committed at least 18.7% of the patients committed at least
one violent act during either the first or sec- one violent act during either the first or sec-
ond 10-week follow-up period), this meant ond 10-week follow-up period), this meant
that the cut-off for the low-risk category that the cut-off for the low-risk category
was 9% violent and the cut-off for the was 9% violent and the cut-off for the
high-risk category was 37% violent. high-risk category was 37% violent.
The ICT contained three iterations (Fig. The ICT contained three iterations (Fig.
1). In the first iteration, the tree classified 1). In the first iteration, the tree classified
429 of the 939 subjects (45.7%) into either 429 of the 939 subjects (45.7%) into either
the high- or the low-risk categories. In the the high- or the low-risk categories. In the
second iteration, the tree classified as high- second iteration, the tree classified as high-
or low-risk 167 (32.7%) of the 510 subjects or low-risk 167 (32.7%) of the 510 subjects
who were not classified into either high- or who were not classified into either high- or
low-risk groups at the end of iteration 1. In low-risk groups at the end of iteration 1. In
the third iteration, the tree classified as the third iteration, the tree classified as
high- or low-risk 86 of the 343 subjects high- or low-risk 86 of the 343 subjects
(25.1%) who were unclassified at the end (25.1%) who were unclassified at the end
of iteration 2. At the end of iteration 3, of iteration 2. At the end of iteration 3,
no further groups could be classified as no further groups could be classified as
high- or low-risk, given the parameters of high- or low-risk, given the parameters of
the model we had set (e.g. no group with the model we had set (e.g. no group with
fewer than 50 cases); 257 subjects (27.4% fewer than 50 cases); 257 subjects (27.4%
of the total sample) remained unclassified. of the total sample) remained unclassified.
The final ICT contained 15 contingent risk The final ICT contained 15 contingent risk
factors that formed 11 risk groups (four factors that formed 11 risk groups (four
low-risk groups, accounting for 50.9% of low-risk groups, accounting for 50.9% of
the total sample; three high-risk groups, ac- the total sample; three high-risk groups, ac-
counting for 21.7% of the total sample; and counting for 21.7% of the total sample; and
four unclassified risk groups, accounting four unclassified risk groups, accounting
for 27.4% of the sample). for 27.4% of the sample).
The risk factors displayed in Fig. 1 are The risk factors displayed in Fig. 1 are
defined as follows. defined as follows. Seriousness of prior ar- Seriousness of prior ar-
rests rests was a patient's self-report of the ser- was a patient's self-report of the ser-
iousness of arrests since age 15 years. iousness of arrests since age 15 years.
Motor impulsiveness Motor impulsiveness was measured from was measured from
the motor sub-scale of the Barratt Impul- the motor sub-scale of the Barratt Impul-
siveness Scale (Barratt, 1994). siveness Scale (Barratt, 1994). Father used Father used
drugs drugs was a self-report question on whether was a self-report question on whether
the patient's father ever used drugs exces- the patient's father ever used drugs exces-
sively. sively. Recent violent fantasies Recent violent fantasies was was
316 316
Fig. 1 Fig. 1 Clinically useful iterative classification tree. Clinically useful iterative classification tree.
^ ^ ^, Low violence risk; , high violence risk; ^ ^ ^, Low violence risk; , high violence risk;
. . . . . . . .
, unclassified. , unclassified.
ACTUARI AL TOOL FOR AS S ES S ING VIOLENCE RI SK ACTUARI AL TOOL F OR AS S ES SING VIOLENCE RI S K
measured by the Schedule of Imagined Vio- measured by the Schedule of Imagined Vio-
lence (Grisso lence (Grisso et al et al, 2000). , 2000). Major disorder Major disorder
without substance abuse without substance abuse refers to a diag- refers to a diag-
nosis of any major mental disorder without nosis of any major mental disorder without
any co-occurring substance abuse diag- any co-occurring substance abuse diag-
nosis, as reached by research clinicians nosis, as reached by research clinicians
using the DSMIIIR Checklist. using the DSMIIIR Checklist. Legal sta- Legal sta-
tus tus was the initial status for the baseline was the initial status for the baseline
hospitalisation, as recorded in hospital ad- hospitalisation, as recorded in hospital ad-
mission records. mission records. Schizophrenia Schizophrenia was the was the
diagnosis reached by research clinicians diagnosis reached by research clinicians
using the DSMIIIR Checklist. using the DSMIIIR Checklist. Anger re- Anger re-
action action was measured by a short version of was measured by a short version of
the Behavioural Subscale of the Novaco the Behavioural Subscale of the Novaco
Anger Scale (Novaco, 1994). Anger Scale (Novaco, 1994). Employed Employed
was a self-report question regarding the pa- was a self-report question regarding the pa-
tient's paid full- or part-time employment tient's paid full- or part-time employment
status in the two months prior to hospital status in the two months prior to hospital
admission. admission. Recent violence Recent violence was a self-report was a self-report
of violence in the two months prior to of violence in the two months prior to
hospital admission. hospital admission. Loss of consciousness Loss of consciousness
referred to a self-report of any loss of referred to a self-report of any loss of
consciousness due to head injury. consciousness due to head injury. Parents Parents
fought fought was a self-report by the patient that was a self-report by the patient that
his or her parents engaged in physical fights his or her parents engaged in physical fights
with one another when the patient was with one another when the patient was
growing up. (A complete list of the ques- growing up. (A complete list of the ques-
tions comprising these risk factors is avail- tions comprising these risk factors is avail-
able from the first author upon request.) able from the first author upon request.)
Receiver operating characteristic Receiver operating characteristic
To assess the predictive accuracy of the ac- To assess the predictive accuracy of the ac-
tuarial model produced by this method and tuarial model produced by this method and
to facilitate further comparisons of our re- to facilitate further comparisons of our re-
sults with other research on violence risk sults with other research on violence risk
assessment, we used a receiver operating assessment, we used a receiver operating
characteristic (ROC) analysis (Gardner characteristic (ROC) analysis (Gardner et et
al al, 1996; Quinsey , 1996; Quinsey et al et al, 1998). The statistic , 1998). The statistic
used to summarise the analysis is the area used to summarise the analysis is the area
under the ROC curve, which corresponds under the ROC curve, which corresponds
to the probability that a randomly selected to the probability that a randomly selected
violent patient will have been assessed by violent patient will have been assessed by
the risk assessment tool as higher risk than the risk assessment tool as higher risk than
a randomly selected non-violent patient a randomly selected non-violent patient
(Swets, 1988). The area under the ROC (Swets, 1988). The area under the ROC
curve for the 11 risk groups presented in curve for the 11 risk groups presented in
Fig. 1 is 0.80 ( Fig. 1 is 0.80 (P P5 50.001). The distribution 0.001). The distribution
of cases that were violent or not violent of cases that were violent or not violent
during the follow-up as a function of the during the follow-up as a function of the
low- and high-risk cut-offs used to generate low- and high-risk cut-offs used to generate
the ICT is presented in Table 2. the ICT is presented in Table 2.
Bootstrapping Bootstrapping
We did not cross-validate the ICT. Cross- We did not cross-validate the ICT. Cross-
validation of a risk assessment model validation of a risk assessment model
requires estimating the model on a subset requires estimating the model on a subset
of the data and validating the model on of the data and validating the model on
the rest. As noted by Gardner the rest. As noted by Gardner et al et al
(1996), however, cross-validation ``wastes (1996), however, cross-validation ``wastes
information that ought to be used estimat- information that ought to be used estimat-
ing the model'' (p 43). For this reason, ing the model'' (p 43). For this reason,
bootstrapping (Mooney & Duval, 1993) bootstrapping (Mooney & Duval, 1993)
has become a widely used analytical strat- has become a widely used analytical strat-
egy for estimating the shrinkage to be egy for estimating the shrinkage to be
expected when a model is generalised to a expected when a model is generalised to a
sample other than the one on which it sample other than the one on which it
was estimated. In conducting such an ana- was estimated. In conducting such an ana-
lysis, 1000 bootstrapped samples were lysis, 1000 bootstrapped samples were
drawn from the original data set. Table 3 drawn from the original data set. Table 3
presents the 95% confidence intervals for presents the 95% confidence intervals for
each of the 11 risk groups in the ICT, in each of the 11 risk groups in the ICT, in
order of decreasing risk. The ranges of these order of decreasing risk. The ranges of these
intervals indicate how the ICT is likely to intervals indicate how the ICT is likely to
perform on other similar samples. perform on other similar samples.
DISCUSSION DISCUSSION
We have sought to increase the utility of an We have sought to increase the utility of an
actuarial method for real-world clinical actuarial method for real-world clinical
decision-making by applying the newly decision-making by applying the newly
developed ICT method (Steadman developed ICT method (Steadman et al et al, ,
2000) to a set of violence risk factors com- 2000) to a set of violence risk factors com-
monly available in clinical records or cap- monly available in clinical records or cap-
able of being assessed routinely in clinical able of being assessed routinely in clinical
practice. We have shown that the ICT par- practice. We have shown that the ICT par-
titioned 72.6% of a sample of discharged titioned 72.6% of a sample of discharged
psychiatric patients into one of two cate- psychiatric patients into one of two cate-
gories with regard to their risk of violence gories with regard to their risk of violence
to others during the first 20 weeks after dis- to others during the first 20 weeks after dis-
charge. One category consisted of groups charge. One category consisted of groups
whose rates of violence were no more than whose rates of violence were no more than
half the base rate of the total patient sam- half the base rate of the total patient sam-
ple (i.e. ple (i.e. 4 49% violent). The other category 9% violent). The other category
consisted of groups whose rates of violence consisted of groups whose rates of violence
were at least twice the base rate of the total were at least twice the base rate of the total
patient sample (i.e. patient sample (i.e. 5 537% violent). The 37% violent). The
actually observed rates of violence in the actually observed rates of violence in the
low- and high-risk categories were 5% low- and high-risk categories were 5%
and and 45%, respectively. The prevalence of 45%, respectively. The prevalence of
violence within individual risk groups violence within individual risk groups
within the low- and high-risk categories within the low- and high-risk categories
varied from 2.7% to 52.7% (Table 3). varied from 2.7% to 52.7% (Table 3).
The ICT left 27.4% of the total sample The ICT left 27.4% of the total sample
unclassified, meaning that it could find no unclassified, meaning that it could find no
combination of risk factors that allowed combination of risk factors that allowed
these patients to be classified into either a these patients to be classified into either a
low- or a high-risk group. The violence rate low- or a high-risk group. The violence rate
for the unclassified category was 24.1%. for the unclassified category was 24.1%.
Clinical illustrations Clinical illustrations
Illustrating the use of the ICT may be help- Illustrating the use of the ICT may be help-
ful. A clinician evaluating a patient's risk of ful. A clinician evaluating a patient's risk of
violence using the ICT presented in Fig. 1. violence using the ICT presented in Fig. 1.
would first ask the patient about the ser- would first ask the patient about the ser-
iousness of his or her prior arrests. If the iousness of his or her prior arrests. If the
patient stated that he or she had previously patient stated that he or she had previously
been arrested for a violent crime, the clini- been arrested for a violent crime, the clini-
cian would then inquire into whether the cian would then inquire into whether the
patient recently had been fantasising about patient recently had been fantasising about
being violent. If the patient responded affir- being violent. If the patient responded affir-
matively to this second question, he or she matively to this second question, he or she
at that point would be placed in the high at that point would be placed in the high
violence risk category. More specifically, violence risk category. More specifically,
the patient would be placed in risk group the patient would be placed in risk group
B, a group in which approximately 53% B, a group in which approximately 53%
of the patients are expected to commit a of the patients are expected to commit a
violent act in the next several months. violent act in the next several months.
If, on the other hand, the patient denied If, on the other hand, the patient denied
having violent fantasies, the clinician would having violent fantasies, the clinician would
then indicate whether the patient had a then indicate whether the patient had a
diagnosis of schizophrenia. If the patient diagnosis of schizophrenia. If the patient
did have such a diagnosis, he or she at that did have such a diagnosis, he or she at that
point would be placed in the low violence point would be placed in the low violence
risk category. More specifically, the patient risk category. More specifically, the patient
317 317
Table 2 Table 2 Distribution of violent and non-violent cases using two thresholds Distribution of violent and non-violent cases using two thresholds
Observed Observed Classification Classification
Low Low5 59% 9% Unclassified Unclassified High High 4 437% 37% Total Total
Not violent Not violent
Violent Violent
Total Total
456 456
22 22
478 478
195 195
62 62
257 257
112 112
92 92
204 204
763 763
176 176
939 939
Table 3 Table 3 Bootstrapped 95% confidence intervals Bootstrapped 95% confidence intervals
for the ICTrisk groups for the ICTrisk groups
Risk group Risk group % violent in risk % violent in risk
group group
95% confidence 95%confidence
interval interval
B B
F F
A A
G G
J J
I I
K K
C C
E E
H H
D D
52.7 52.7
41.8 41.8
39.2 39.2
34.4 34.4
28.2 28.2
18.9 18.9
10.9 10.9
7.4 7.4
6.8 6.8
4.7 4.7
2.7 2.7
41.0^63.8 41.0^63.8
31.3^52.5 31.3^52.5
26.2^52.4 26.2^52.4
22.5^46.1 22.5^46.1
20.6^35.8 20.6^35.8
8.2^29.4 8.2^29.4
3.6^18.2 3.6^18.2
1.6^13.2 1.6^13.2
1.3^12.1 1.3^12.1
0.2^8.8 0.2^8.8
0.5^4.9 0.5^4.9
MONAHAN E T AL MONAHAN E T AL
would be placed in risk group E, a group in would be placed in risk group E, a group in
which approximately 7% of the patients which approximately 7% of the patients
are expected to commit a violent act in are expected to commit a violent act in
the next several months. (For other studies the next several months. (For other studies
finding rates of violence to be lower among finding rates of violence to be lower among
patients with schizophrenia than among patients with schizophrenia than among
patients with other, primarily personality patients with other, primarily personality
disorder, diagnoses, see: Gardner disorder, diagnoses, see: Gardner et al et al, ,
1996; Quinsey 1996; Quinsey et al et al, 1998; Wallace , 1998; Wallace et al et al, ,
1998.) 1998.)
Comparative predictive accuracy Comparative predictive accuracy
We have demonstrated here that the ICT We have demonstrated here that the ICT
method may be adapted for clinical use. method may be adapted for clinical use.
The method does not require unavailable The method does not require unavailable
or costly-to-gather data in order to charac- or costly-to-gather data in order to charac-
terise the risk of violence. Rather, risk fac- terise the risk of violence. Rather, risk fac-
tors usually found in patient files, or tors usually found in patient files, or
capable of routine assessment, are all that capable of routine assessment, are all that
are required for the ICT to function. The are required for the ICT to function. The
predictive accuracy of the ICT using a re- predictive accuracy of the ICT using a re-
duced set of 106 clinically feasible risk fac- duced set of 106 clinically feasible risk fac-
tors from the MacArthur Violence Risk tors from the MacArthur Violence Risk
Assessment Study (an area under the ROC Assessment Study (an area under the ROC
curve of 0.80) is comparable to the predic- curve of 0.80) is comparable to the predic-
tive accuracy that we reported (Steadman tive accuracy that we reported (Steadman et et
al al, 2000) for risk assessment using the , 2000) for risk assessment using the
expanded set of 134 risk factors (an area expanded set of 134 risk factors (an area
under the ROC curve of 0.82). under the ROC curve of 0.82).
Violence risk assessment software Violence risk assessment software
Although the contingent nature of the risk Although the contingent nature of the risk
factors identified in Fig. 1 may appear too factors identified in Fig. 1 may appear too
intricate for use in clinical practice, the util- intricate for use in clinical practice, the util-
ity of the ICT model would be enhanced ity of the ICT model would be enhanced
greatly with the aid of software. Software greatly with the aid of software. Software
would facilitate the assessment of an indivi- would facilitate the assessment of an indivi-
dual patient by guiding the clinician to ask dual patient by guiding the clinician to ask
only those questions required to assess risk. only those questions required to assess risk.
We are in the process of developing such We are in the process of developing such
software. software.
ACKNOWLEDGEMENTS ACKNOWLEDGEMENTS
This research was supported by the Research Net- This research was supported by the Research Net-
work on Mental Health and the Law of the John D. work on Mental Health and the Law of the John D.
and Catherine T. MacArthur Foundation and by and Catherine T. MacArthur Foundation and by
NIMH grant R01 49696. We are indebted to the NIMH grant R01 49696. We are indebted to the
members of the Network (Shirley S. Abrahamson, members of the Network (Shirley S. Abrahamson,
Richard J. Bonnie, Pamela S. Hyde, Stephen J. Morse, Richard J. Bonnie, Pamela S. Hyde, Stephen J. Morse,
Paul Slovic and David B.Wexler) and to Steven Banks Paul Slovic and David B.Wexler) and to Steven Banks
and RoumenVesselinov. and RoumenVesselinov.
REFERENCES REFERENCES
American Psychiatric Association American Psychiatric Association (1989) (1989) Diagnostic Diagnostic
and Statistical Manual of Mental Disorders and Statistical Manual of Mental Disorders (3rd edn, (3rd edn,
revised) (DSM^III ^R).Washington, DC: American revised) (DSM^III ^R).Washington, DC: American
Psychiatric Association. Psychiatric Association.
Appelbaum, P. S., Robbins, P. C. & Roth, L. H. Appelbaum, P. S., Robbins, P. C. & Roth, L. H. (1999) (1999)
Adimensional approach to delusions: comparison across A dimensional approach to delusions: comparison across
delusion types and diagnoses. delusion types and diagnoses. American Journal of American Journal of
Psychiatry Psychiatry, , 156 156, 1938^1943. , 1938^1943.
Barratt, E. Barratt, E. (1994) (1994) Impulsiveness and aggression. In Impulsiveness and aggression. In
Violence and Mental Disorder: Developments in Risk Violence and Mental Disorder: Developments in Risk
Assessment Assessment (eds J. Monahan & H. Steadman), pp. 61^79. (eds J. Monahan & H. Steadman), pp. 61^79.
Chicago, IL: University of Chicago Press. Chicago, IL: University of Chicago Press.
Borum, R. Borum, R. (1996) (1996) Improving the clinical practice of Improving the clinical practice of
violence risk assessment: technology, guidelines, and violence risk assessment: technology, guidelines, and
training. training. American Psychologist American Psychologist, , 51 51, 945^956. , 945^956.
Breiman, L., Freidman, J., Olshen, R., Breiman, L., Freidman, J., Olshen, R., et al et al (1984) (1984)
Classification and RegressionTrees Classification and RegressionTrees. Pacific Grove, CA: . Pacific Grove, CA:
Wadsworth and Brooks/Cole. Wadsworth and Brooks/Cole.
Cohen, S., Kamarck, T. & Mermelstein, R. Cohen, S., Kamarck, T. & Mermelstein, R. (1983) (1983) A A
global measure of perceived stress. global measure of perceived stress. Journal of Health and Journal of Health and
Social Behavior Social Behavior, , 24 24, 385^396. , 385^396.
Estroff, S. & Zimmer, C. Estroff, S. & Zimmer, C. (1994) (1994) Social networks, Social networks,
social support, and violence among persons with severe, social support, and violence among persons with severe,
persistent mental illness. In persistent mental illness. In Violence and Mental Disorder: Violence and Mental Disorder:
Developments in Risk Assessment Developments in Risk Assessment (eds J. Monahan & H. (eds J. Monahan & H.
Steadman), pp. 259^295. Chicago, IL: University of Steadman), pp. 259^295. Chicago, IL: University of
Chicago Press. Chicago Press.
Folstein, M., Folstein, S. & McHugh, P. Folstein, M., Folstein, S. & McHugh, P. (1975) (1975) Mini- Mini-
mental state: a practical method for grading the mental state: a practical method for grading the
cognitive state of patients for the clinician. cognitive state of patients for the clinician. Journal of Journal of
Psychiatric Research Psychiatric Research, , 12 12, 189^198. , 189^198.
Gardner, W., Lidz, C., Mulvey, E. P., Gardner, W., Lidz, C., Mulvey, E. P., et al et al (1996) (1996) A A
comparison of actuarial methods for identifying comparison of actuarial methods for identifying
repetitively violent patients with mental illness. repetitively violent patients with mental illness. Law and Law and
Human Behavior Human Behavior, , 20 20, 35^48. , 35^48.
Grisso, T., Davis, J.,Vesselinov, R., Grisso, T., Davis, J.,Vesselinov, R., et al et al (2000) (2000) Violent Violent
thoughts and violent behavior following hospitalization thoughts and violent behavior following hospitalization
for mental disorder. for mental disorder. Journal of Consulting and Clinical Journal of Consulting and Clinical
Psychology Psychology, in press. , in press.
Hare, R. Hare, R. (1998) (1998) The Hare PCL^R: some issues The Hare PCL^R: some issues
concerning its use and misuse. concerning its use and misuse. Legal and Criminological Legal and Criminological
Psychology Psychology, , 3 3, 99^119. , 99^119.
Hart, S., Cox, D. & Hare, R. Hart, S., Cox, D. & Hare, R. (1995) (1995) The Hare The Hare
Psychopathy Checklist: Screening Version Psychopathy Checklist: Screening Version. North . North
Tonawanda, NY: Multi-Health Systems. Tonawanda, NY: Multi-Health Systems.
318 318
CLINICAL IMPLICATIONS CLINICAL IMPLICATIONS
& &
The classification tree approach to violence risk assessment proposed here The classification tree approach to violence risk assessment proposed here
improves on traditional approaches by explicitly acknowledging that violence is an improves on traditional approaches by explicitly acknowledging that violence is an
outcome reached by multiple routes. outcome reachedby multiple routes.
& &
Employing two cut-off scores ^ one for identifying high-riskcases and one for Employing two cut-off scores ^ one for identifying high-riskcases and one for
identifyinglow-riskcases ^ is morerealistic than attempting to classify all patients as identifyinglow-riskcases ^ is morerealistic thanattempting to classify all patients as
to the risk of violence. to the risk of violence.
& & Aclinically useful actuarial approach for assessing the riskof violence among acute Aclinically useful actuarial approach for assessing the riskof violence among acute
psychiatric patients nowexists. psychiatric patients nowexists.
LIMITATIONS LIMITATIONS
& &
This is a clinical study of violence among people hospitalised for mental disorders, This is a clinical study of violence among people hospitalised for mental disorders,
not an epidemiological study of violence among people with mental disorders in the not an epidemiological study of violence among people with mental disorders in the
general population. general population.
& & The extent to which the accuracy of the actuarial tool developedhere generalises The extent to which the accuracy of the actuarial tool developedhere generalises
to other types of clinical setting (e.g. forensic hospitals) is unknown. to other types of clinical setting (e.g. forensic hospitals) is unknown.
& & The proposed tool is specifically designed for assessing the riskof violence; efforts The proposed tool is specifically designed for assessing the riskof violence; efforts
to manage the risk of violence will require additional data. to manage the risk of violence will require additional data.
JOHN MONAHANPhD, School of Law, University of Virginia, Charlottesville,VA 22903; HENRY J. JOHN MONAHANPhD, School of Law, University of Virginia, Charlottesville,VA 22903; HENRY J.
STEADMAN, PhD, PAMELAC. ROBBINS, BA, Policy Research Associates, Delmar, NY12054; ERIC SILVER, STEADMAN, PhD, PAMELAC. ROBBINS, BA, Policy Research Associates, Delmar, NY12054; ERIC SILVER,
PhD, Department of Sociology, The Pennsylvania State University, College Park, PA16802; PAUL S. PhD, Department of Sociology, The Pennsylvania State University, College Park, PA16802; PAUL S.
APPELBAUM, MD, THOMAS GRISSO, PhD, Department of Psychiatry, University of Massachusetts Medical APPELBAUM, MD, THOMAS GRISSO, PhD, Department of Psychiatry, University of Massachusetts Medical
Center,Worcester, MA 01655; EDWARD P. MULVEY, PhD, LORENH. ROTH, MD,Western Psychiatric Institute Center,Worcester, MA 01655; EDWARD P. MULVEY, PhD, LORENH. ROTH, MD,Western Psychiatric Institute
and Clinic, University of Pittsburgh School of Medicine, Pittsburgh, PA15213, USA and Clinic, University of Pittsburgh School of Medicine, Pittsburgh, PA15213, USA
Correspondence: John Monahan, PhD, School of Law,University of Virginia, 580 Massie Road, Correspondence: John Monahan, PhD, School of Law,University of Virginia, 580 Massie Road,
Charlottesville,VA 22903-1789,USA. Tel: (804) 924 3632; Fax: (804) 982 2845; e-mail: Charlottesville,VA 22903-1789,USA. Tel: (804) 924 3632; Fax: (804) 982 2845; e-mail:
jmonahan jmonahan@ @law5.law.virginia.edu law5.law.virginia.edu
(First received19 May 1999, final revision 3 December 1999, accepted 7 December 1999) (First received19 May 1999, final revision 3 December 1999, accepted 7 December 1999)
ACTUARI AL TOOL FOR AS S ES S ING VIOLENCE RI SK ACTUARI AL TOOL F OR AS S ES SING VIOLENCE RI S K
Hollingshead, A. & Redlich, F. Hollingshead, A. & Redlich, F. (1958) (1958) Social Class and Social Class and
Mental Illness Mental Illness. NewYork: Guilford Press. . NewYork: Guilford Press.
Janca, A. & Helzer, J. Janca, A. & Helzer, J. (1990) (1990) DSM^III ^RCriteria DSM^III ^R Criteria
Checklist. Checklist. DIS Newsletter DIS Newsletter, , 7 7, 17. , 17.
Levy, D. E., Caronna, J. J., Singer, B. H., Levy, D. E., Caronna, J. J., Singer, B. H., et al et al (1985) (1985)
Predicting outcome fromhypoxic^ischemic coma. Predicting outcome from hypoxic^ischemic coma.
Journal of the American Medical Association Journal of the American Medical Association, , 253 253, ,
1420^1426. 1420^1426.
Lidz, C., Mulvey, E. P. & Gardner, W. Lidz, C., Mulvey, E. P. & Gardner, W. (1993) (1993) The The
accuracy of predictions of violence to others. accuracy of predictions of violence to others. Journal of Journal of
the American Medical Association the American Medical Association, , 269 269, 1007^1011. , 1007^1011.
Monahan, J., Appelbaum, P., Mulvey, E., Monahan, J., Appelbaum, P., Mulvey, E., et al et al (1994) (1994)
Ethical and legal duties in conducting research on Ethical and legal duties in conducting research on
violence: lessons fromthe MacArthur Risk Assessment violence: lessons fromthe MacArthur Risk Assessment
Study. Study. Violence and Victims Violence and Victims, , 8 8, 380^390. , 380^390.
Mooney, C. & Duval, R. Mooney, C. & Duval, R. (1993) (1993) Bootstrapping: A Bootstrapping: A
Nonparametric Approach to Statistical Inference Nonparametric Approach to Statistical Inference. Newbury . Newbury
Park, CA: Sage. Park, CA: Sage.
Novaco, R. Novaco, R. (1994) (1994) Anger as a risk factor for violence Anger as a risk factor for violence
among the mentally disordered. In among the mentally disordered. In Violence and Mental Violence and Mental
Disorder: Developments in Risk Assessment Disorder: Developments in Risk Assessment (eds J. (eds J.
Monahan & H. Steadman), pp. 21^59. Chicago, IL: Monahan & H. Steadman), pp. 21^59. Chicago, IL:
University of Chicago Press. University of Chicago Press.
Overall, J. Overall, J. (1988) (1988) The Brief Psychiatric Rating Scale The Brief Psychiatric Rating Scale
(BPRS): recent developments in ascertainment and (BPRS): recent developments in ascertainment and
scaling. scaling. Psychopharmacology Bulletin Psychopharmacology Bulletin, , 24 24, 97^99. , 97^99.
Pescosolido, B., Monahan, J., Link, B., Pescosolido, B., Monahan, J., Link, B., et al et al (1999) (1999)
The public's view of the competence, dangerousness and The public's view of the competence, dangerousness and
need for legal coercion among persons with mental need for legal coercion among persons with mental
illness. illness. American Journal of Public Health American Journal of Public Health, , 89 89, 1339^1345. , 1339^1345.
Quinsey,V., Harris, G., Rice, M., Quinsey,V., Harris, G., Rice, M., et al et al (1998) (1998) Violent Violent
Offenders: Appraising and Managing Risk. Offenders: Appraising and Managing Risk.Washington, Washington,
DC: American Psychological Association. DC: American Psychological Association.
Reed, J. Reed, J. (1997) (1997) Risk assessment and clinical risk Risk assessment and clinical risk
management: the lessons from recent inquiries. management: the lessons fromrecent inquiries. British British
Journal of Psychiatry Journal of Psychiatry, , 170 170 (suppl. 32), 4^7. (suppl. 32), 4^7.
SPSS SPSS (1993) (1993) SPSS for Windows CHAID SPSS for Windows CHAID (Release 6.0). (Release 6.0).
Chicago, IL: SPSS Inc. Chicago, IL: SPSS Inc.
Steadman, H. J., Monahan, J., Appelbaum, P. S., Steadman, H. J., Monahan, J., Appelbaum, P. S., et al et al
(1994) (1994) Designing a new generation of risk assessment Designing a new generation of risk assessment
research. In research. In Violence and Mental Disorder: Developments Violence and Mental Disorder: Developments
in Risk Assessment in Risk Assessment (eds J. Monahan & H. Steadman), (eds J. Monahan & H. Steadman),
pp. 297^318. Chicago, IL: University of Chicago Press. pp. 297^318. Chicago, IL: University of Chicago Press.
_ _ , Mulvey, E. P., Monahan, J., , Mulvey, E. P., Monahan, J., et al et al (1998) (1998) Violence Violence
by people discharged from acute psychiatric inpatient by people discharged from acute psychiatric inpatient
facilities and by others in the same neighborhoods. facilities and by others in the same neighborhoods.
Archives of General Psychiatry Archives of General Psychiatry, , 55 55, 1^9. , 1^9.
_ _ , Silver, E., Monahan, J., , Silver, E., Monahan, J., et al et al (2000) (2000) A A
classification tree approach to the development of classification tree approach to the development of
actuarial violent risk assessment tools. actuarial violent risk assessment tools. Law and Human Law and Human
Behavior Behavior, , 24 24, 83^100. , 83^100.
Swets, J. Swets, J. (1988) (1988) Measuring the accuracy of diagnostic Measuring the accuracy of diagnostic
systems. systems. Science Science, , 240 240, 1285^1293. , 1285^1293.
Wallace, C., Mullen, P., Burgess, P., Wallace, C., Mullen, P., Burgess, P., et al et al (1998) (1998)
Serious criminal offending and mental disorder. Case Serious criminal offending and mental disorder. Case
linkage study. linkage study. British Journal of Psychiatry British Journal of Psychiatry, , 172 172, 477^484. , 477^484.
319 319