Skip to main content
PLOS ONE logoLink to PLOS ONE
. 2012 Feb 16;7(2):e31727. doi: 10.1371/journal.pone.0031727

A Novel Approach to Determining Violence Risk in Schizophrenia: Developing a Stepped Strategy in 13,806 Discharged Patients

Jay P Singh 1, Martin Grann 2, Paul Lichtenstein 3, Niklas Långström 2, Seena Fazel 1,*
Editor: Leonardo Fontenelle4
PMCID: PMC3280996  PMID: 22359622

Abstract

Clinical guidelines recommend that violence risk be assessed in schizophrenia. Current approaches are resource-intensive as they employ detailed clinical assessments of dangerousness for most patients. An alternative approach would be to first screen out patients at very low risk of future violence prior to more costly and time-consuming assessments. In order to implement such a stepped strategy, we developed a simple tool to screen out individuals with schizophrenia at very low risk of violent offending. We merged high quality Swedish national registers containing information on psychiatric diagnoses, socio-demographic factors, and violent crime. A cohort of 13,806 individuals with hospital discharge diagnoses of schizophrenia was identified and followed for up to 33 years for violent crime. Cox regression was used to determine risk factors for violent crime and construct the screening tool, the predictive validity of which was measured using four outcome statistics. The instrument was calibrated on 6,903 participants and cross-validated using three independent replication samples of 2,301 participants each. Regression analyses resulted in a tool composed of five items: male sex, previous criminal conviction, young age at assessment, comorbid alcohol abuse, and comorbid drug abuse. At 5 years after discharge, the instrument had a negative predictive value of 0.99 (95% CI = 0.98–0.99), meaning that very few individuals who the tool screened out (n = 2,359 out of original sample of 6,903) were subsequently convicted of a violent offence. Screening out patients who are at very low risk of violence prior to more detailed clinical assessment may assist the risk assessment process in schizophrenia.

Introduction

Although schizophrenia is consistently associated with an increased risk of violence compared with general population controls in different countries and using various definitions of violence, most individuals with schizophrenia are not dangerous [1], [2]. A recent population study estimated rates of violent crime to be approximately 10–15% after diagnosis [1], and cohort studies have reported rates between 8–14% [3][6]. Despite this, current treatment guidelines in the US and UK recommend that violence risk be assessed for all individuals diagnosed with schizophrenia [7], [8], and over 100 instruments have been developed to assist in the prediction of such behaviour in general and psychiatric populations [9], [10]. Available instruments focus on identifying those patients with schizophrenia who are at highest risk of violence and typically require several hours to administer, placing a considerable burden on the resources of mental health services [11], [12].

One approach that could improve the risk assessment process is to develop instruments to screen out patients who are at very low risk of future violence that can be used prior to in-depth assessments of dangerousness (in contrast to current one-step methods that administer costly risk assessments to all patients without initial screening). This stepped approach is similar to that used by many diagnostic screening tools in physical medicine, such as the use of mammography or prostatic specific antigen screening for breast or prostate cancer, respectively, prior to biopsy [13], and allows for clinical resources, including more detailed clinical risk assessment, to be targeted at those more likely to need intervention. In addition, using such an approach would assist in risk management by identifying individuals who could be targeted for treatment in those who are not screened out.

The objective of the present study was to develop a screening tool to identify individuals with schizophrenia who are at very low risk of violence after hospital discharge. We developed and cross-validated the instrument using a cohort of 13,806 discharged patients with schizophrenia, with a potential follow-up of 33 years. Rather than simply turning risk assessment on its head (i.e., substituting factors such as previous criminal conviction with no previous conviction), the tool was statistically developed to maximise sensitivity (i.e. identifying all those individuals not at very low risk) for use as part of a stepped strategy. We aimed to create a tool composed of routinely available factors that was potentially scalable as it does not require specific training or additional costs. These properties have been identified as important for any instrument aiming to be clinically relevant [14].

Methods

Study protocol

The Standards for Reporting of Diagnostic Accuracy Studies (STARD) Statement [15], a 25-item checklist of reporting characteristics, was followed.

Data registries

Data was collected from the following nationwide population-based registers in Sweden: The Hospital Discharge Register (HDR; held at the National Board of Health and Welfare), the Migration Register (Statistics Sweden), the Cause of Death Register (National Board of Health and Welfare), the National Crime Register (National Council for Crime Prevention), Education Registers (Statistics Sweden), and the Multi-Generation Register (Statistics Sweden). Merging these datasets was possible as all national registers use residents' (including immigrants) same 10-digit personal identification numbers. An independent government organisation (Statistics Sweden) merged the registers and, upon assigning each participant a novel case number, destroyed the coding sheet that linked case numbers and personal identification numbers. This anonymisation made informed consent unnecessary. The Ethics Committee at the Karolinska Institutet approved this protocol (2005/174/31/4).

Participants

Using the HDR, all individuals aged 15 years (the age of criminal responsibility) and older who had been admitted to hospital for assessment and/or treatment in Sweden and had been discharged between January 1, 1973 and December 31, 2004 with a diagnosis of schizophrenia were identified. The HDR is a high quality register with less than 1% of hospital discharges missing personal identification numbers [16]. Diagnoses of schizophrenia were made using the International Classification of Diseases (ICD-8: 295; ICD-9: 295; ICD-10: F20). Only those individuals who had been diagnosed with schizophrenia on at least two occasions (N = 13,806) were included to improve specificity [17].

Participants were followed until their first conviction or until they were censored due to emigration (data obtained from the Migration Register), death (data obtained from the Cause of Death Register), or the end of the follow-up period (December 31, 2004).

Calibration and cross-validation samples

The population cohort was randomly divided into two samples of 6,903 participants. One of the samples was used to develop the screening tool and the other to cross-validate it. As the use of regression to develop risk instruments has been criticised for exaggerating predictive validity estimates in calibration samples [18], [19], the cross-validation sample was randomly split into three independent subsamples, each consisting of 2,301 participants. Cross-validating the tool using these subsamples provided the opportunity to assess shrinkage effects.

Definition of violence

Violent conviction data was extracted from the National Crime Register for the years 1973–2004. The National Crime Register is a high quality register, missing only 0.05% of personal identification numbers between 1988–2000 [16]. Violent crime was defined as having received a conviction for homicide, assault, robbery, arson, any sexual offence, illegal threats, or intimidation. The inclusion of noncontact offences in the operational definition of violence is consistent with previous risk assessment research [20], [21], including studies that have specifically investigated the association between schizophrenia and criminal behaviour [1]. Nonviolent offences such as fraud and perjury were not included, as they were perceived to be of less clinical interest than more serious behavioural outcomes and as they are not referenced in current treatment guidelines for schizophrenia [7], [8]. Participants were coded as either having or not having a conviction for a violent crime during follow-up. Conviction data accurately reflects resolved criminality in this population as plea-bargaining is not permitted in Sweden and as individuals are convicted as guilty regardless of mental illness.

Outcome measures

Four outcome statistics were used to measure predictive validity: negative predictive value (NPV), positive predictive value (PPV), diagnostic odds ratio (DOR), and area under the curve (AUC). These statistics are commonly used outcome measures in the medical diagnostic literature [22].

The NPV is the proportion of individuals who are predicted by a tool not to offend who do not offend, whereas the PPV represents the proportion of individuals who are predicted to offend who actually offend. Both statistics are base rate dependent. As the ability to accurately classify individuals as being at risk of violence in psychiatric populations is highly dependent upon base rate [23], this may be considered a strength.

The DOR is the ratio of the odds of a true positive relative to the odds of a false positive. In addition to being independent of the base rate of offending, the DOR is also useful in that researchers and clinicians are familiar with the concept of an odds ratio, making the statistic easier to comprehend for non-specialists [24].

The receiver operating characteristic (ROC) curve plots an instrument's true positive rate against its false positive rate across score thresholds. The area under this curve (AUC) is the probability that a randomly selected offender has a higher test score than a randomly selected non-offender [25]. The statistic can be considered an index of how well a tool discriminates between offenders and non-offenders. As with the DOR, the AUC is not base rate dependent.

Risk factors

We prioritised the inclusion of risk factors that could be scored using routinely collected and accessible file information. These variables included male sex [26], [27], previous criminal conviction [27], age at assessment [27], comorbid alcohol abuse [28], and comorbid drug (non-alcohol) abuse [28].

Information on sex was extracted from the HDR. Previous criminal conviction referred to any crime (violent or not) before hospital discharge, and was gathered from the National Crime Register. The age at which participants were discharged from hospital with a second diagnosis of schizophrenia ranged from 15 to 54 years (Median = 26; Interquartile Range = 22–33). We used the Chi-squared Automatic Interaction Detector (CHAID) method adjusted by time at risk to identify the age cut-off which best classified participants as being at risk of violent conviction [29]. Data on participants' hospital admissions during the years 1973–2004 were extracted from the HDR and examined for evidence of principal or comorbid diagnoses of alcohol abuse (ICD-8: 303; ICD-9: 303, 305.1; ICD-10: F10, except ×.5) and drug abuse (ICD-8: 304; ICD-9: 304, 305.9; ICD-10: F11-F19, except ×.5). This information was used as a marker of substance abuse comorbidity. A sensitivity analysis was also conducted for the following illicit drug categories: opiate, sedative, stimulant, hallucinogen, and polysubstance abuse.

Secondary risk factors tested included low level of education [30], having a parent who was convicted of a violent offence [31], and having a parent who was diagnosed with alcohol abuse [32]. These were considered secondary, as such information may not be routinely available. One study estimated that school records were utilized in approximately 15% of clinicians conducting risk assessments in adults [33]. Participants' highest level of completed education was collated from the Education Registers and a binary comparison was made between those who had completed compulsory school (a nine year comprehensive school for children aged 7 to 16) and those who had not.

We linked parents to the National Crime Register using the Multi-Generational Register to extract data on whether individuals had a father or mother who had previously been convicted of a violent crime. In addition, parents were linked to the HDR to extract data on diagnoses of alcohol abuse.

Developing the screening tool

Analyses took place in two stages: in the first stage, the routinely accessible variables were combined into a five-item unit scored model (i.e., a model where the presence of a risk factor is scored as +1 and the absence as +0) and regressed with violent conviction as the dependent variable using Cox's proportional-hazards method [34]. The rates of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) were calculated for the resulting tool at each risk score at 1, 2, and 5 years follow-up. Participants who were not at risk for 1, 2, or 5 years (respectively) due to emigration, death, or end of follow-up were excluded from these calculations.

When selecting a cut-off score, the relative costs of false negatives and false positives were taken into consideration. The societal and political costs associated with screening out a patient who would go on to commit a violent crime were considered to outweigh the costs associated with conducting more detailed risk assessments for patients who would not go on to commit a violent offence. Therefore, the ideal cut-off point was identified as the risk score that maximised sensitivity while balancing specificity. Using this cut-off, the instrument's NPV, PPV, and DOR were calculated at each length of follow-up, as was AUC using receiver operating characteristic analysis.

In the second stage, each secondary variable was regressed with violent conviction as the dependent variable and ranked in order of their resulting hazard ratios. Variables were added one at a time to the primary model and, provided the added factor independently accounted for variation in the dependent variable, the revised tool's predictive validity was assessed at 1, 2, and 5 years follow-up. Effect estimates were compared with those produced by the five-item model at the same length of follow-up using 95% confidence intervals (CI). Additional comparisons were made using the χ2 test of differences between proportions, Breslow and Day's χ2 test of homogeneity between odds ratios [35], and Hanley and McNeil's z test for differences in AUC [36]. This process was repeated for each additive iteration.

Cross-validation

The version of the screening tool that produced the highest rates of predictive validity using the fewest number of items was then cross-validated using the three replication subsamples. The effect estimates produced by the tool in each independent subsample at 1, 2, and 5 years follow-up were compared with those produced by the calibration sample at the same length of follow-up to examine evidence of shrinkage.

Analyses were conducted using SPSS 17.0.1 for Windows [37], STATA/IC 10.1 for Windows [38], and MedCalc 11.3.8.0 for Windows [39].

Results

Description of the samples

The population cohort consisted of 13,806 individuals with two or more hospital discharge diagnoses of schizophrenia during January 1, 1973 and December 31, 2004 (Table 1). The cohort was randomly divided into a calibration sample (n = 6,903) and three independent cross-validation subsamples (n = 2,301 for each).

Table 1. Descriptive characteristics of the calibration and cross-validation samples of a violence screening tool for individuals with hospital discharge diagnoses of schizophrenia.

Sample
Domain Variable Calibration sample (n = 6903) Cross-validation sample 1 (n = 2301) Cross-validation sample 2 (n = 2301) Cross-validation sample 3 (n = 2301) Adjusted HRa(95% CI)
Demographic factors Male sex, n (%) 4392 (63.6) 1480 (64.3) 1504 (65.4) 1515 (65.8) 3.3 (2.7–4.0)
Age at assessment (in years), mean (SD) 28.9 (7.2) 28.9 (7.2) 29.2 (7.2) 29.2 (7.3) 1.9 (1.6–2.3)
Non-completion of compulsory school, n (%) 2005 (29.1) 660 (28.7) 679 (29.5) 692 (30.1) 1.4 (1.2–1.6)
Individual factors Previous criminal conviction, n (%) 2806 (40.7) 899 (39.1) 974 (42.3) 978 (42.5) 3.3 (2.8–3.8)
Alcohol abuse comorbidity, n (%) 1078 (15.6) 353 (15.3) 353 (15.3) 331 (14.4) 2.9 (2.5–3.4)
Drug (non-alcohol) abuse comorbidity, n (%) 1151 (16.7) 360 (15.6) 389 (16.9) 357 (15.5) 3.5 (3.1–4.0)
Familial factors Father convicted of a violent offence, n (%) 127 (1.8) 41 (1.8) 39 (1.7) 35 (1.5) 2.3 (1.7–3.2)
Mother convicted of a violent offence, n (%) 17 (0.2) 7 (0.3) 6 (0.3) 7 (0.3) 2.8 (1.4–5.3)
Father alcohol abuse comorbidity, n (%) 539 (7.8) 188 (8.2) 163 (7.1) 147 (6.4) 1.4 (1.1–1.8)
Mother alcohol abuse comorbidity, n (%) 179 (2.6) 56 (2.4) 60 (2.6) 68 (3.0) 1.8 (1.3–2.5)
Base rate of violent convictionb n (%) 887 (12.9) 309 (13.4) 321 (14.0) 275 (12.0)

Note: HR = hazard ratio; CI = confidence interval; SD = standard deviation.

a

Adjusted by age at assessment and sex using participants from the calibration sample.

b

Between January 1, 1973 and December 31, 2004.

Calibrating the screening tool

Using the CHAID method, we identified 32 years as the age cut-off that best classified participants as being at risk of violent conviction for the “young age at assessment” variable.

Cox regression found that the five routinely accessible factors (i.e., male sex, previous criminal conviction, young age at assessment, comorbid alcohol abuse, and comorbid drug abuse) were significant independent predictors of the incidence of violent conviction (Table 1).

The TP, FP, TN, and FN rates were calculated for the five-item tool at each risk score at 1, 2, and 5 years follow-up. The optimal cut-off score was identified as +2 (Table 2). Of the 2,359 participants who scored below this threshold, 2,353 were not convicted of a violent offence within 1 year after hospital discharge (NPV = 0.99; 95% CI = 0.99–1.00). The screening tool's PPV at 1 year after discharge was 0.01 (95% CI = 0.01–0.02), the DOR was 3.79 (95% CI = 1.65–8.72), and the AUC was 0.67 (95% CI = 0.66–0.68). There was a trend towards PPVs, DORs, and AUCs increasing over 2 and 5 years follow-up, while NPVs remained at 99% (Table 3). To investigate the instrument's performance with different base rates, we measured PPVs and NPVs at 1, 2, and 5 years follow-up at base rates of 0%, 2%, 4%, 6%, 8%, and 10% (Table 4). While PPVs increased markedly with increasing base rates, NPVs remained above 95%.

Table 2. Rates of true positives, true negatives, false positives, and false negatives by risk score for the calibration sample of a violence screening tool for individuals with hospital discharge diagnoses of schizophrenia.

Length of follow-up Risk score True positives True negatives False positives False negatives Sensitivity Specificity
1 year (n = 6645) 0 47 0 6598 0 1.00 0.00
1 46 604 5994 1 0.98 0.09
2* 41 2353 4245 6 0.87 0.36
3 26 4493 2105 21 0.55 0.68
4 14 5750 848 33 0.30 0.87
5 3 6380 218 44 0.06 0.97
2 years (n = 6407) 0 93 0 6314 0 1.00 0.00
1 90 573 5741 3 0.97 0.09
2* 83 2245 4069 10 0.89 0.36
3 58 4300 2014 35 0.62 0.68
4 29 5505 809 64 0.31 0.87
5 9 6104 210 84 0.10 0.97
5 years (n = 5666) 0 224 0 5442 0 1.00 0.00
1 216 475 4967 8 0.96 0.09
2* 202 1961 3481 22 0.90 0.36
3 138 3762 1680 86 0.62 0.69
4 64 4776 666 160 0.29 0.88
5 15 5261 181 209 0.07 0.97

Note: Participants from the calibration sample were excluded if they were not at risk for 1, 2, or 5 years (respectively).

*Cut-off score.

Table 3. Comparison of outcome measures calculated during calibration and cross-validation of a violence screening tool for individuals with hospital discharge diagnoses of schizophrenia.

Outcome measure
Length of follow-up Sample NPV (95% CI) PPV (95% CI) DOR (95% CI) AUC (95% CI)
1 year Calibration sample (n = 6645) 0.99 (0.99–1.00) 0.01 (0.01–0.02) 3.79 (1.65–8.72) 0.67 (0.59–0.74)
Cross-validation sample 1 (n = 2205) 0.99 (0.99–1.00) 0.01 (0.01–0.02) 5.16 (1.33–20.04) 0.74 (0.62–0.85)
Cross-validation sample 2 (n = 2205) 0.99 (0.99–1.00) 0.01 (0.01–0.02) 2.07 (0.73–5.93) 0.66 (0.54–0.78)
Cross-validation sample 3 (n = 2190) 0.99 (0.99–1.00) 0.01 (0.01–0.02) 5.14 (1.34–19.93) 0.62 (0.53–0.71)
2 years Calibration sample (n = 6407) 0.99 (0.99–1.00) 0.02 (0.02–0.03) 4.58 (2.40–8.74) 0.69 (0.64–0.74)
Cross-validation sample 1 (n = 2136) 0.99 (0.99–1.00) 0.02 (0.02–0.03) 17.89 (3.08–103.80) 0.75 (0.67–0.82)
Cross-validation sample 2 (n = 2111) 0.99 (0.98–1.00) 0.02 (0.02–0.03) 2.56 (1.09–6.03) 0.67 (0.58–0.75)
Cross-validation sample 3 (n = 2111) 0.99 (0.99–1.00) 0.02 (0.02–0.03) 3.12 (1.24–7.82) 0.65 (0.57–0.73)
5 years Calibration sample (n = 5666) 0.99 (0.98–0.99) 0.06 (0.05–0.06) 5.17 (3.33–8.03) 0.69 (0.66–0.73)
Cross-validation sample 1 (n = 1875) 0.99 (0.98–0.99) 0.06 (0.05–0.06) 4.10 (2.12–7.91) 0.68 (0.63–0.74)
Cross-validation sample 2 (n = 1879) 0.98 (0.97–0.99) 0.05 (0.05–0.06) 3.08 (1.63–5.81) 0.68 (0.62–0.74)
Cross-validation sample 3 (n = 1868) 0.98 (0.97–0.99) 0.05 (0.04–0.05) 2.21 (1.23–3.95) 0.66 (0.60–0.72)

Note: NPV = negative predictive value; PPV = positive predictive value; DOR = diagnostic odds ratio; AUC = area under the curve; CI = confidence interval. Participants from the calibration and cross-validation samples were excluded if they were not at risk for 1, 2, or 5 years (respectively). All AUCs were significantly higher than chance. Test statistics reported no significant differences between the effect estimates produced by the calibration and cross-validation samples. The instrument's PPV increased significantly over time in both the calibration and the cross-validation samples. Differences in DOR and AUC were non-significant despite evidence of non-overlapping confidence intervals.

Table 4. A comparison of the positive and negative predictive values for a violence screening tool for individuals with hospital discharge diagnoses of schizophrenia across different base rates of violent conviction.

Base rate of violent conviction
Length of follow-up Outcome measure 0% 2% 4% 6% 8% 10%
1 year (n = 6645) NPV (95% CI) 0.99 (0.99–1.00) 0.99 (0.99–1.00) 0.99 (0.98–0.99) 0.98 (0.97–0.98) 0.97 (0.97–0.98) 0.96 (0.96–0.97)
PPV (95% CI) 0.01 (0.00–0.01) 0.03 (0.01–0.07) 0.05 (0.03–0.08) 0.08 (0.06–0.11) 0.11 (0.08–0.13) 0.13 (0.11–0.15)
2 years (n = 6407) NPV (95% CI) 0.99 (0.99–1.00) 0.99 (0.99–1.00) 0.99 (0.98–0.99) 0.98 (0.97–0.98) 0.97 (0.97–0.98) 0.97 (0.96–0.97)
PPV (95% CI) 0.01 (0.00–0.01) 0.03 (0.01–0.07) 0.05 (0.03–0.09) 0.08 (0.06–0.11) 0.11 (0.09–0.13) 0.13 (0.11–0.16)
5 years (n = 5666) NPV (95% CI) 0.99 (0.99–1.00) 0.99 (0.99–1.00) 0.99 (0.98–0.99) 0.98 (0.97–0.98) 0.98 (0.97–0.98) 0.97 (0.96–0.97)
PPV (95% CI) 0.01 (0.00–0.01) 0.03 (0.01–0.07) 0.06 (0.04–0.09) 0.08 (0.06–0.11) 0.11 (0.09–0.13) 0.14 (0.11–0.16)

Note: PPV = positive predictive value; NPV = negative predictive value; CI = confidence interval. Values based on participants from the calibration sample. Participants were excluded if they were not at risk for 1, 2, or 5 years (respectively).

Univariate analyses also found that a number of secondary variables were significantly associated with the incidence of violent conviction (Table 1). However, we found no evidence of increased predictive validity when low level of education, parental conviction for a violent offence, or parental alcohol abuse comorbidity were added as items (results not shown). In addition, in a sensitivity analysis, no evidence of incremental predictive validity was found if one of opiate, sedative, stimulant, hallucinogen, or polysubstance abuse was included on the simple screening tool instead of any drug abuse.

Cross-validation

Statistical tests revealed no significant differences between the effect estimates produced by the five-item instrument in the calibration and cross-validation samples at each length of follow-up, suggesting no clear shrinkage effects (Table 3). Of the 2,288 participants in the replication samples who were classified as low risk, 2,280 did not go on to be convicted of a violent offence within 1 year of hospital discharge (NPV = 0.99; 95% CI = 0.99–1.00). Similar to the calibration sample, there was a trend towards PPVs, DORs, and AUCs increasing over time, while NPVs remained at or above 98% at 2 and 5 years of follow-up (Table 3).

Discussion

Using data collected from Swedish national registers over three decades, we investigated violence risk in 13,806 patients with schizophrenia in order to develop a simple screening tool that could accurately identify individuals who would not be convicted of a violent offence after hospital discharge. The instrument was composed of five routinely available risk factors: male sex, previous criminal conviction, young age at assessment, comorbid alcohol abuse, and comorbid drug abuse. We found that the tool could be accurately used to make “rule out” decisions (i.e., identifying who will not go on to violently offend), suggesting potential utility as a screening tool that could be used as part of a stepped approach to risk assessment. In this approach, very low risk patients are screened out, allowing for more detailed and resource-intensive assessment and management on the remaining (or screen positive) patients. The reported tool requires no training to use and hence is potentially scalable.

Inclusion of clinical override

The screening tool should complement the role of clinical judgement in violence risk assessment by identifying patients for whom detailed assessment may not be needed. Nevertheless, a number of risk factors that this tool does not measure may have clinical significance and more research is warranted to determine their value. For example, associations have been found between violence and non-adherence to medication [40], lack of insight [41], and psychotic symptoms such as persecutory delusions and command hallucinations [42]. In certain cases, clinicians may consider such factors as highly relevant (e.g., if they were associated with previous violence). Therefore, we have included a clinical override option in our coding sheet (Figure 1).

Figure 1. Proposed violence screening tool for individuals with schizophrenia.

Figure 1

Comparison with previous research

While several violence risk assessment instruments have been developed for general psychiatric populations [43][45], no tools to our knowledge have been designed as part of a stepped strategy for individuals with a specific psychiatric diagnosis [10]. In relation to the latter, as epidemiological investigations have shown that the base rate of violence varies by diagnosis [46] and existing tools may perform differently across diagnostic groups [10], general screening approaches are likely to lose sensitivity and specificity. Some evidence suggests that developing risk assessment tools for more specific populations may improve predictive validity [47]. One such instrument is an actuarial tool designed to predict assault in individuals diagnosed with psychotic disorders [48]. The calibration study for this instrument, a two-year clinical trial examining different models of community care in UK inner cities, found that demographic information was helpful in risk prediction [48], [49]. However. unlike the screening tool developed in the present study, this measure was designed to identify high risk individuals and hence produced a higher rate of false negatives (13% at 2 year follow-up compared with less than 1% in the present study).

The current screening instrument's success in screening out very low risk patients may be attributed to two sources. First, the tool was statistically developed so that it only includes risk factors that account for variance independently, and the selected cut-off maximises sensitivity and minimises the number of false negative predictions. Second, each of the risk factors has been proposed to lie on the causal pathway to violence. For example, increased levels of testosterone may predispose male patients with schizophrenia to aggression [50]. Further, there is evidence to suggest that the presence of the Met allele in the catechol-O-methyltransferase (COMT) Val158Met polymorphism, which has recently been put forth as an important neurobiological risk factor for violence in schizophrenia [51], increases violence risk solely in men [52]. Young age may increase violence risk, as aging reduces testosterone levels in both sexes [53] and is associated with improved emotion regulation [54]. Criminal history and diagnoses of alcohol or drug abuse may be markers of violence risk because they may reflect an antisocial lifestyle with an increased likelihood of having antisocial peers, procriminal attitudes, unemployment, exposure to destabilizers, and lack of adherence to treatment [55]. In addition, alcohol or drug use at the time of the violent offence may have acted as a disinhibitor. Despite this, this study was not designed to test causal mechanisms, and the screening tool clearly does not represent a causal model of violence risk.

To our knowledge, the simple screen presented in this report is the first risk assessment tool to have been designed to predict the likelihood of community violence in a specific diagnostic group, to have been constructed for use in identifying low as opposed to high risk patients, and to have been developed using a register-based sample of several thousand individuals, allowing for the calculation of more precise effect estimates than previous work. The calibration and cross-validation samples for the screening tool were several orders of magnitude larger than development studies for currently available actuarial risk assessment tools [56], [57].

Implications

When used as part of a stepped approach to violence risk assessment, screen out tools can potentially save psychiatric services considerable resources although future research will have to test this in practice. Recent UK and US surveys have found that instruments which employ structured clinical judgement, where clinicians use empirically-based risk and protective factors to guide their predictions as to whether an individual will be violent, are amongst the most commonly used tools [58], [59]. With the time involved in administering such tools (familiarising oneself with the patient's case, collecting information from multiple sources to score items, conducting interviews, scoring the tool, and making a clinical judgement regarding risk level) and the costs involved (attending training sessions, purchasing tool manuals, and paying for each coding sheet), violence risk assessment, as often currently conducted, costs mental health services significant amounts of time and money. In a recent survey of forensic mental health professionals, Viljoen and colleagues [33] reported that using a structured tool to conduct a risk assessment takes an average of 15 hours and costs service providers an average of $100 USD per hour to complete. Therefore, screening tools that require little training, that are based on items scored using routinely available information, and can reduce caseload are potentially attractive. It should be noted, however, that patients with schizophrenia who are screened out will continue to need mental health resources to treat their illness. Therefore, while screening tools such as that developed in the present study may be of use in identifying patients who do not need management of their violence risk, they should not be used to ration treatment.

The results of the present investigation also suggest that reporting a single effect size does not provide an adequate picture of a risk assessment tool's predictive validity. Global effect estimates such as the AUC and the DOR do not measure the relative utility of an instrument in making “rule in” decisions (i.e., identifying who will go on to violently offend) versus “rule out” decisions (i.e., identifying who will not go on to violently offend). Had only AUCs or DORs have been reported for the simple tool in this study, different conclusions could have emerged about how the tool performed. AUCs below 0.70 may have been interpreted as evidence that the screening tool lacked any utility as a violence risk assessment instrument [60], whereas DORs above 5 may have been interpreted as evidence that the screen performed well in identifying both high and low risk patients [24]. It was only by examining the PPVs and NPVs of the tool that a clear picture emerged of how the instrument performed. Future studies investigating the predictive validity of forensic risk assessment tools should consider reporting a global effect estimate (e.g., AUC or DOR) as well as at least one “rule in” and “rule out” effect size (e.g., PPV and NPV, respectively).

Limitations

There are several limitations of the present investigation. First, we relied on hospital data to identify schizophrenia cases. Therefore, our screen is applicable only to patients discharged from hospital, although other research has estimated that a minority of individuals in Sweden who have schizophrenia would not have been hospitalised over a 30-year period [61]. By excluding individuals with only a single discharge diagnosis, we will not have included all hospitalised individuals with schizophrenia. However, by including only those individuals with two diagnoses of schizophrenia, our sample had the advantage of diagnostic specificity [17]. That is, participants were less likely to have other diagnoses such as drug-induced psychosis or bipolar disorder. Related work has shown similar base rates of violence in those with one or more diagnoses of schizophrenia [1], so it is possible that the screen is generalisable to all hospitalised patients with the disorder. Further, as the prevalence [62], [63] and disability-adjusted life years [64] of individuals with schizophrenia and the rate of violent crime [65] in Sweden are similar to Western Europe and other regions, our findings may have some generalisability to other nations.

A second limitation was that the positive predictive values produced by the screening instrument were low. That is, patients who were not screened out by the simple tool were rarely convicted of violent crimes (i.e., false positives). This is likely a consequence of the low base rate of violence in individuals with schizophrenia and supports the assumption of the stepped approach that individuals who score highly on our tool should not be considered as moderate or high risk of violence, but rather as warranting a more comprehensive risk assessment. In this context, false positive predictions are not problematic, as a low-risk patient who is not screened out would receive what is currently treatment-as-usual: an in-depth clinically-based risk assessment. These findings highlight the challenge of identifying individuals at high risk of violence in populations with low base rates [66], and also the need for more research on improving structured risk instruments, possibly using other variables that we were unable to test using register-based measures.

A third limitation of the present investigation was that evidence of minimal shrinkage effects may have been due to our calibration and cross-validation samples having been selected from the same population, which had a low overall base rate of violence. Also, variables were coded the same way for both the construction and replication samples. Future work will have to test the predictive validity of the screening tool in different settings (e.g., non-hospitalised patients) and outcomes (e.g., violence not reported to the police).

A fourth limitation is that the screening tool does not discriminate in young men, who would all screen positive. Therefore, it may be that the screen provides the most clinical utility in women and older men. Future research could consider more discriminating tools in young men.

Future Directions

Before the stepped approach to risk assessment could be considered in clinical practice, it would need further validation, including clinical studies in other countries to assess the generalisability of the present study's findings. These studies should be prospective and include both economic analyses (i.e., comparisons of resources consumed prior to implementation and after) and qualitative analyses to explore clinicians' perceptions of this novel approach.

In addition, as clinicians may be interested in tools that predict more than one adverse outcome [67], future studies may wish to investigate the utility of the screening instrument in the prediction of other adverse outcomes such as suicide. In support, a systematic review has found male sex, criminal history, young age, and substance abuse to be significant risk factors for suicide in schizophrenia [68][70].

To what extent the five factors included on the screening tool are specific to schizophrenia also needs further investigation. For two factors, it is unlikely: the age cut-off in this instrument (32 years) is older than the median age of offending in general populations [71], and rates of substance abuse comorbidity are substantially higher in patients with schizophrenia than general population samples [72].

Conclusion

The aim of the present study was to develop a simple instrument to identify those individuals diagnosed with schizophrenia who are at very low risk of violence after hospital discharge. We aimed to design a tool that used routinely available information and was straightforward to administer. With reasonably high rates of accuracy, the use of five items, no complicated weighting algorithm, and an easy to interpret classification system (screen out vs. continue risk assessment), our findings suggest that using a stepped strategy in which very low risk patients are screened out prior to in-depth risk assessment may assist in improving the quality of violence risk assessment in schizophrenia.

Acknowledgments

We are grateful to Eva Carlström, who extracted the data.

Footnotes

Competing Interests: The authors have declared that no competing interests exist.

Funding: Support for this study was provided by The Swedish Research Council (Medicine). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

  • 1.Fazel S, Långström N, Hjern A, Grann M, Lichtenstein P. Schizophrenia, substance abuse, and violent crime. JAMA. 2009;301:2016–2023. doi: 10.1001/jama.2009.675. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Walsh E, Buchanan A, Fahy T. Violence and schizophrenia: Examining the evidence. Br J Psychiatry. 2002;180:490–495. doi: 10.1192/bjp.180.6.490. [DOI] [PubMed] [Google Scholar]
  • 3.Brennan PA, Mednick SA, Hodgins S. Major mental disorders and criminal violence in a Danish birth cohort. Arch Gen Psychiatry. 2000;57:494–500. doi: 10.1001/archpsyc.57.5.494. [DOI] [PubMed] [Google Scholar]
  • 4.Lindqvist P, Allebeck P. Schizophrenia and crime: A longitudinal follow-up of 644 schizophrenics in Stockholm. Br J Psychiatry. 1990;157:345–350. doi: 10.1192/bjp.157.3.345. [DOI] [PubMed] [Google Scholar]
  • 5.Räsänen P, Tiihonen J, Isohanni M, Rantakallio P, Lehtonen J, et al. Schizophrenia, alcohol abuse, and violent behavior: A 26-year followup study of an unselected birth cohort. Schizophr Bull. 1998;24:437–441. doi: 10.1093/oxfordjournals.schbul.a033338. [DOI] [PubMed] [Google Scholar]
  • 6.Soyka M, Graz C, Bottlender R, Dirschedl P, Schoech H. Clinical correlates of later violence and criminal offences in schizophrenia. Schizophr Res. 2007;94:89–98. doi: 10.1016/j.schres.2007.03.027. [DOI] [PubMed] [Google Scholar]
  • 7.American Psychiatric Association. Practice guidelines for the treatment of patients with schizophrenia. Arlington, VA: American Psychiatric Association; 2004. [Google Scholar]
  • 8.National Institute for Health and Clinical Excellence. Core interventions in the treatment and management of schizophrenia in primary and secondary care. London: National Institute for Health and Clinical Excellence; 2009. [Google Scholar]
  • 9.Singh JP, Fazel S. Forensic risk assessment: A metareview. Crim Justice Behav. 2010;37:965–988. [Google Scholar]
  • 10.Singh JP, Serper M, Reinharth J, Fazel S. Structured assessment of violence risk in schizophrenia and other psychiatric disorders: A systematic review of the validity, reliability, and item content of 10 available instruments. Schizophr Bull. 2011;37:899–912. doi: 10.1093/schbul/sbr093. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Large MM, Ryan CJ, Singh SP, Paton MB, Nielssen OB. The predictive value of risk categorization in schizophrenia. Harv Rev Psychiatry. 2011;19:25–33. doi: 10.3109/10673229.2011.549770. [DOI] [PubMed] [Google Scholar]
  • 12.Murrie DC, Cornell D, McCoy WK. Psychopathy, conduct disorder, and stigma: Does diagnostic labelling influence juvenile probation officer recommendations? Law Hum Behav. 2005;29:323–342. doi: 10.1007/s10979-005-2415-x. [DOI] [PubMed] [Google Scholar]
  • 13.Smith RA, Cokkinides V, Brooks D, Saslow D, Shah M, et al. Cancer screening in the United States, 2011: A review of current American Cancer Society guidelines and issues in cancer screening. CA - Cancer J Clin. 2011;61:8–30. doi: 10.3322/caac.20096. [DOI] [PubMed] [Google Scholar]
  • 14.Webster C, Polvi N. Challenging assessments of dangerousness and risk. In: Ziskin J, editor. Coping with psychiatric and psychological testimony. Marina del Ray, CA: Law and Psychology Press; 1995. [Google Scholar]
  • 15.Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glaziou PP, et al. Towards complete and accurate reporting of studies of diagnostic accuracy: The STARD initiative. BMJ. 2003;326:41–44. doi: 10.1136/bmj.326.7379.41. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Fazel S, Grann M. The population impact of severe mental illness on violent crime. Am J Psychiatry. 2006;163:1397–1403. doi: 10.1176/ajp.2006.163.8.1397. [DOI] [PubMed] [Google Scholar]
  • 17.Reutfors J, Bahmanyar S, Jönsson EG, Ekbom A, Nordström P, et al. Diagnostic profile and suicide risk in schizophrenia spectrum disorder. Schizophr Res. 2010;123:251–256. doi: 10.1016/j.schres.2010.07.014. [DOI] [PubMed] [Google Scholar]
  • 18.Blair PR, Marcus DK, Boccaccini MT. Is there an allegiance effect for assessment instruments? Actuarial risk assessment as an exemplar. Clin Psychol Sci Pract. 2010;15:346–360. [Google Scholar]
  • 19.Pedhazur E. Multiple regression in behavioral research. Orlando: Harcourt Brace; 1997. [Google Scholar]
  • 20.Hanson RK, Morton-Bourgon K. Predictors of sexual recidivism: An updated meta-analysis. Ottawa, ON: Public Works and Government Services Canada; 2004. [Google Scholar]
  • 21.Mills JF, Kroner DG. The effect of discordance among violence and general recidivism risk estimates on predictive accuracy. Crim Behav Ment Health. 2006;16:155–166. doi: 10.1002/cbm.623. [DOI] [PubMed] [Google Scholar]
  • 22.Honest H, Khan KS. Reporting of measures of accuracy in systematic reviews of diagnostic literature. BMC Health Serv Res. 2002;2:1–4. doi: 10.1186/1472-6963-2-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Large MM, Ryan CJ, Nielssen OB. Helpful and unhelpful risk assessment practices. Psychiatr Serv. 2010;61:530. doi: 10.1176/appi.ps.61.5.530. [DOI] [PubMed] [Google Scholar]
  • 24.Glas AS, Lijmer JG, Prins MH, Bonsel GJ, Bossuyt PM. The diagnostic odds ratio: A single indicator of test performance. J Clin Epidemiol. 2003;56:1129–1135. doi: 10.1016/s0895-4356(03)00177-x. [DOI] [PubMed] [Google Scholar]
  • 25.Swets JA, Dawes RM, Monahan J. Psychological science can improve diagnostic decisions. Psychol Sci Public Interest. 2000;1:1–26. doi: 10.1111/1529-1006.001. [DOI] [PubMed] [Google Scholar]
  • 26.Dean K, Walsh E, Morgan C, Demjaha A, Dazzan P, et al. Aggressive behaviour at first contact with services: Findings from the AESOP First Episode Psychosis Study. Psychol Med. 2007;37:547–557. doi: 10.1017/S0033291706008920. [DOI] [PubMed] [Google Scholar]
  • 27.Swanson JW, Swartz MS, Van Dorn RA, Elbogen EB, Wagner HR, et al. A national study of violent behavior in persons with schizophrenia. Arch Gen Psychiatry. 2006;63:490–499. doi: 10.1001/archpsyc.63.5.490. [DOI] [PubMed] [Google Scholar]
  • 28.Fazel S, Gulati G, Linsell L, Geddes JR, Grann M. Schizophrenia and violence: Systematic review and meta-analysis. PLoS Med. 2009;6:e1000120. doi: 10.1371/journal.pmed.1000120. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Kass GV. An exploratory technique for investigating large quantities of categorical data. Appl Stat. 1980;29:119–127. [Google Scholar]
  • 30.Cannon M, Huttunen MO, Tanskanen AJ, Arseneault L, Jones PB, et al. Perinatal and childhood risk factors for later criminality and violence in schizophrenia. Br J Psychiatry. 2002;180:496–501. doi: 10.1192/bjp.180.6.496. [DOI] [PubMed] [Google Scholar]
  • 31.Steadman HJ, Mulvey EP, Monahan J. Violence by people discharged from acute psychiatric inpatient facilities and by others in the same neighborhoods. Arch Gen Psychiatry. 1998;55:393–401. doi: 10.1001/archpsyc.55.5.393. [DOI] [PubMed] [Google Scholar]
  • 32.Pilowsky DJ, Keyes KM, Hasin DS. Adverse childhood events and lifetime alcohol dependence. Am J Public Health. 2009;99:258–263. doi: 10.2105/AJPH.2008.139006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Viljoen JL, McLachlan K, Vincent GM. Assessing violence risk and psychopathy in juvenile and adult offenders: A survey of clinical practices. Assessment. 2010;17:377–395. doi: 10.1177/1073191109359587. [DOI] [PubMed] [Google Scholar]
  • 34.Cox DR. Regression models and life tables. J Roy Stat Soc B Stat Meth. 1972;34:187–220. [Google Scholar]
  • 35.Breslow N, Day N. Statistical methods in cancer research. Lyon: International Agency for Research on Cancer; 1987. [Google Scholar]
  • 36.Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology. 1983;148:839–843. doi: 10.1148/radiology.148.3.6878708. [DOI] [PubMed] [Google Scholar]
  • 37.SPSS Inc. SPSS for Windows: Release 17.0.1. College Station, TX: SPSS Inc; 2009. [Google Scholar]
  • 38.StataCorp. STATA Statistical Software: Release 10.1. Chicago: StataCorp LP; 2007. [Google Scholar]
  • 39.MedCalc Software. MedCalc: Version 11.3.8.0. Mariakerke, Belgium: MedCalc Software; 2010. [Google Scholar]
  • 40.Swartz MS, Swanson JW, Hiday VA, Borum R, Wagner HR, et al. Violence and severe mental illness: The effects of substance abuse and nonadherence to medication. Am J Psychiatry. 1998;155:226–231. doi: 10.1176/ajp.155.2.226. [DOI] [PubMed] [Google Scholar]
  • 41.Buckley PF, Hrouda DR, Friedman L, Noffsinger SG, Resnick PJ, et al. Insight and its relationship to violent behavior in patients with schizophrenia. Am J Psychiatry. 2004;161:1712–1714. doi: 10.1176/appi.ajp.161.9.1712. [DOI] [PubMed] [Google Scholar]
  • 42.Bjørkly S. Psychotic symptoms and violence towards others: A literature review of some preliminary findings. Aggress Viol Behav. 2002;7:617–631. [Google Scholar]
  • 43.Monahan J, Steadman HJ, Appelbaum PS, Robbins PC, Mulvey EP, et al. Developing a clinically useful actuarial tool for assessing violence risk. Br J Psychiatry. 2000;176:312–319. doi: 10.1192/bjp.176.4.312. [DOI] [PubMed] [Google Scholar]
  • 44.Roaldset JO, Hartvig P, Bjørkly S. V-RISK-10: Validation of a screen for risk of violence after discharge from acute psychiatry. Eur Psychiatry. 2011;26:85–91. doi: 10.1016/j.eurpsy.2010.04.002. [DOI] [PubMed] [Google Scholar]
  • 45.Watts D, Bindman J, Slade M, Holloway F, Rosen A, et al. Clinical Assessment of Risk Decision Support (CARDS): The development and evaluation of a feasible violence risk assessment for routine psychiatric practice. J Ment Health. 2004;13:569–581. [Google Scholar]
  • 46.Stuart HL, Arboleda-Florez JE. A public health perspective on violent offenses among persons with mental illness. Psychiatr Serv. 2001;52:654–659. doi: 10.1176/appi.ps.52.5.654. [DOI] [PubMed] [Google Scholar]
  • 47.Singh JP, Grann M, Fazel S. A comparative study of violence risk assessment tools: A systematic review and metaregression analysis of 68 studies involving 25,980 participants. Clin Psychol Rev. 2011;31:499–513. doi: 10.1016/j.cpr.2010.11.009. [DOI] [PubMed] [Google Scholar]
  • 48.Wootton L, Buchanan A, Leese M, Tyrer P, Burns T, et al. Violence in psychosis: Estimating the predictive validity of readily accessible clinical information in a community sample. Schizophr Res. 2008;101:176–184. doi: 10.1016/j.schres.2007.12.490. [DOI] [PubMed] [Google Scholar]
  • 49.Walsh E, Gilvarry C, Samele C, Harvey K, Manley C, et al. Predicting violence in schizophrenia: A prospective study. Schizophr Res. 2004;67:247–252. doi: 10.1016/S0920-9964(03)00091-4. [DOI] [PubMed] [Google Scholar]
  • 50.Pope HG, Kourt EM, Hudson JI. Effects of supraphysiologic doses of testosterone on mood and aggression in normal men: A randomized controlled trial. Arch Gen Psychiatry. 2000;57:133–140. doi: 10.1001/archpsyc.57.2.133. [DOI] [PubMed] [Google Scholar]
  • 51.Soyka M. Neurobiology of aggression and violence in schizophrenia. Schizophr Bull. 2011;37:913–920. doi: 10.1093/schbul/sbr103. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Jones G, Zammit S, Norton N, Hamshere ML, Jones SJ, et al. Aggressive behaviour in patients with schizophrenia is associated with catechol-O-methyltransferase genotype. Br J Psychiatry. 2001;179:351–355. doi: 10.1192/bjp.179.4.351. [DOI] [PubMed] [Google Scholar]
  • 53.Harman SM, Metter EJ, Tobin JB, Pearson J, Blackman MR. Longitudinal effects of aging on serum total and free testosterone levels in healthy men. J Clin Endocrinol Metab. 2001;86:724–731. doi: 10.1210/jcem.86.2.7219. [DOI] [PubMed] [Google Scholar]
  • 54.Hariri AR, Bookheimer SY, Mazziotta JC. Modulating emotional responses: Effects of a neocortical network on the limbic system. Neuroreport. 2000;11:43–48. doi: 10.1097/00001756-200001170-00009. [DOI] [PubMed] [Google Scholar]
  • 55.Andrews DA, Bonta J. The psychology of criminal conduct (4th ed.) Newark, NJ: LexisNexis; 2006. [Google Scholar]
  • 56.Andrews DA, Bonta J. LSI-R: The Level of Service Inventory – Revised. Toronto: Multi-Health Systems; 1995. [Google Scholar]
  • 57.Quinsey VL, Harris GT, Rice ME, Cormier CA. Violent offenders: Appraising and managing risk. Washington, DC: American Psychological Association; 2006. [Google Scholar]
  • 58.Archer RP, Buffington-Vollum JK, Stredny RV, Handel RW. A survey of psychological test use patterns among forensic psychologists. J Pers Assess. 2006;87:84–94. doi: 10.1207/s15327752jpa8701_07. [DOI] [PubMed] [Google Scholar]
  • 59.Khiroya R, Weaver T, Maden T. Use and perceived utility of structured violence risk assessments in English medium secure forensic units. Psychiatrist. 2009;33:129–132. [Google Scholar]
  • 60.Sjöstedt G, Grann M. Risk assessment: What is being predicted by actuarial prediction instruments? Int J Forensic Ment Health. 2002;1:179–183. [Google Scholar]
  • 61.Hansson L, Vinding HR, Mackeprang T, Sourander A, Werdelin G, et al. Comparison of key worker and patient assessment of needs in schizophrenic patients living in the community: A Nordic multicentre study. Acta Psychiatr Scand. 2001;103:45–51. doi: 10.1034/j.1600-0447.2001.00083.x. [DOI] [PubMed] [Google Scholar]
  • 62.Saha S, Chant D, Welham J, McGrath J. A systematic review of the prevalence of schizophrenia. PLoS Med. 2005;2:413–433. doi: 10.1371/journal.pmed.0020141. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 63.Wittchen HU, Jacobi F. Size and burden of mental disorders in Europe: A critical review and appraisal of 27 studies. Eur Neuropsychopharmacol. 2005;15:357–376. doi: 10.1016/j.euroneuro.2005.04.012. [DOI] [PubMed] [Google Scholar]
  • 64.World Health Organization. The global burden of disease: 2004 update. Geneva: WHO; 2008. [Google Scholar]
  • 65.Dolmén L. Brottsligheten I olika länder. [Criminality in different countries] Stockholm: National Council for Crime Prevention; 2001. [Google Scholar]
  • 66.Szmukler G. The mathematics of risk assessment for serious violence. Psychiatrist. 2001;25:359. [Google Scholar]
  • 67.Webster CD, Martin ML, Brink J, Nicholls TL, Desmarais SL. Manual for the Short-Term Assessment of Risk and Treatability (START) (Version 1.1) Port Coquitlam, Canada: BC Mental Health & Addiction Services; 2009. [Google Scholar]
  • 68.Gupta S, Black DW, Arndt S, Hubbard WC, Andreasen NC. Factors associated with suicide attempts among patients with schizophrenia. Psychiatr Serv. 1998;49:1353–1355. doi: 10.1176/ps.49.10.1353. [DOI] [PubMed] [Google Scholar]
  • 69.Hawton K, Sutton L, Haw C, Sinclair J, Deeks JJ. Schizophrenia and suicide: Systematic review of risk factors. Br J Psychiatry. 2005;187:9–20. doi: 10.1192/bjp.187.1.9. [DOI] [PubMed] [Google Scholar]
  • 70.Taiminen T, Huttunen J, Heilä H, Henriksson M, Isometsä E, et al. The Schizophrenia Suicide Risk Scale (SSRS): Development and initial validation. Schizophr Res. 2001;47:2–3. doi: 10.1016/s0920-9964(00)00126-2. [DOI] [PubMed] [Google Scholar]
  • 71.Farrington DP. Key results from the first forty years of the Cambridge study in delinquent development. In: Thornberry TP, Krohn MD, editors. Taking stock of delinquency: An overview of findings from contemporary longitudinal studies. New York: Kluwer Academic; 2004. [Google Scholar]
  • 72.Green AI, Drake RE, Brunette MF, Noordsy DL. Schizophrenia and co-occurring substance use disorder. Am J Psychiatry. 2007;164:402–408. doi: 10.1176/ajp.2007.164.3.402. [DOI] [PubMed] [Google Scholar]

Articles from PLoS ONE are provided here courtesy of PLOS

RESOURCES