Violence Risk Assessment Tools Used in Forensic and Acute Psychiatry in North America: A Scoping Review

Abstract

Violence in psychiatric settings presents a significant risk to patients, staff, and society at large. With over 400 risk assessment tools available globally, their applications and the risks they assess vary, allowing for diverse use in different situations. This scoping review investigated the risk management tools utilized in North America’s forensic psychiatry and acute psychiatric units, aiming to identify which ones are mainly used. A comprehensive search was conducted across PubMed, Embase, and PsycINFO databases, following PRISMA Guidelines, covering the literature from their inception date until 2023. Criteria for study inclusion required a focus on risk management tool use in forensic or acute psychiatric settings, originality (original studies, case reports, or systematic reviews), and a North American context. Out of 3059 identified studies, 40 were thoroughly analyzed. Commonly used risk assessment scales include HARM-FV, eHARM-FV, HCR-20, PCL-R, START, BVC, and DASA, with their reliability varying by the clinical context and the assessed population. The review highlights the heterogeneous application of static and dynamic scales across clinical settings, underscoring a need for more precise tools to improve risk assessments in forensic psychiatry, signaling a call for the development and validation of more sophisticated assessment instruments.

1. Introduction

1.1. Violence in Psychiatric Settings

Violence is a reality across various healthcare settings, but managing it is particularly important in forensic psychiatry to facilitate the patient’s rehabilitation [1]. The presence of violence in psychiatric environments poses risks not only to the patients themselves but also to the staff and the wider society [2]. These risks materialize when the potential for agitation or harm is overlooked, potentially prolonging the patient’s reintegration into society, and increasing the likelihood of harm [3,4]. Furthermore, it can affect the productivity of staff members and contribute to burnout [5]. Notably, a recent qualitative study on registered nurses reported that those working in acute care psychiatric units face an elevated risk and often encounter violent and aggressive behavior from patients [6]. Even shift workers are especially at risk of encountering physical violence [7]. While the prevalence of violence in psychiatric wards varies greatly, the literature is consistent in reporting that future studies should prioritize examining the early stages of aggression like agitation, as well as factors conducive to preventing aggression such as ward and staff dynamics, with potential implications for staffing and ward management practices [8,9].

1.2. Risk Factors When Assessing for Violence

Various factors can contribute to predicting violence in patients. These include environmental influences like exposure to violence, poverty, and unstable living conditions; a personal or family history of violence; certain mental health diagnoses, which may or may not be compounded by substance use; and traits such as a lack of empathy or impulsivity [10,11,12,13,14]. Additional patient-related factors to consider are gender, with males and younger individuals being more prone to violence, noncompliance with treatment, and access to weapons, among others [14]. A patient’s background, including a history of fire-setting, cruelty to animals, age at first arrest, and the number of arrests, is also relevant when assessing violence [14].

Both static and dynamic indicators play essential roles in risk assessment tools. Dynamic risk factors, which can change with intervention, must meet three criteria: they must precede and elevate the risk of violence, be capable of change, and have a causal relationship with violence risk, meaning their increase (or decrease) can predict a corresponding change in violence risk [15,16].

1.3. Risk Assessment Tools

Guidelines around the globe recommend structured assessment tools like the Historical Clinical and Risk Management (HCR-20), Psychopathy Checklist Revised Screening Version (PCL-SV), and the Hare Psychopathy Checklist-Revised (PCL-R) for evaluating risk [17]. Currently, there are over 400 risk assessment tools available internationally [18]. Depending on the specific situation and risk being assessed, various risk management instruments can be utilized. For instance, certain scales are designed for assessing acute risk within a 24 h period, such as the Dynamic Appraisal of Situation Aggression-Inpatient Version (DASA-IV), while others, like the Short-Term Assessment of Risk and Treatability (START), are geared towards short-term evaluation [19,20]. These tools are invaluable for forensic psychiatrists and clinicians, enabling them to make well-informed decisions about the management and treatment of individuals in their care. They offer the capability to predict the risk of violence over short or long periods, thereby guiding appropriate preventive measures. Although there is existing literature on various tools, there is a gap in providing an overview of the tools specifically utilized within the North American context. Such an exploration of the literature is important for effectively utilizing the reported tools and for further developing tools tailored to the psychiatric challenges encountered by patients and staff in North America. The recent literature stresses the need for better data regarding risk assessment tools in the context of forensic psychiatry [21].

1.4. Objectives of the Study

This scoping review aims to examine all the risk management tools currently in use within forensic psychiatry and acute psychiatric contexts across North America. The objective is to report metrics on their effectiveness in the North American contexts to evaluate their use in these clinical settings, given their significance for the wellbeing of patients, medical personnel, and society. It is hypothesized that a wide variety of clinical appraisal tools will be identified across these contexts, and they will vary in terms of their effectiveness.

2. Materials and Methods

2.1. Search Strategies

A systematic search was carried out in the electronic databases of Pubmed, Embase, and PsycINFO from 2010 until 2023 to assess the contemporary literature. The search was performed by using inclusive keywords for the fields of psychometric tools (i.e., psychometrics), risk evaluation (i.e., violence, aggression), and risk evaluation in forensic or acute psychiatric contexts (i.e., psychiatric department, hospital, psychiatric). Other studies were found by using additional records that were identified through cross-referencing. Furthermore, Appendix A contains the complete electronic search strategy and the Supplementary Material S1 contains the PRISMA for scoping review checklist. This search method was developed by the main author of this article and a mental health-specialized librarian working at the Institut universitaire en santé mentale de Montréal. The search was accomplished by MAR, OLCH, and AH in 2023. Searches were restricted to English or French language sources, and, at first, there were no limitations about the setting, the date, or the geographical location of the studies found.

2.2. Study Eligibility

For studies to be considered for inclusion in the analysis, they were required to meet the following criteria: (1) the focus of the article should be on the utilization of a risk management tool, (2) it must involve patients within forensic psychiatry or acute psychiatry settings, (3) the publication needs to be an original study, case report, or systematic literature review, (4) the research should have been conducted in North America, and (5) the article must be written in either English or French. Additionally, (6) the publication date of all articles must be from 2010 onwards to encompass the most recent literature on the subject.

2.3. Data Extraction

MAR and OLCH extracted data using a Microsoft Excel (version 16.0) form, which was then cross-validated by AH and SBP. The information gathered included the type of risk assessment tool used, the number of items for each tool, participant numbers, study location, key indicators of interest from each study, and the overall conclusions. These data were compiled and included in the document.

3. Results

3.1. Description of Selected Studies

Our systematic review focused on articles that evaluated violence risk scales within forensic psychiatry or acute psychiatric settings in North America. Initially, we found 3059 articles related to the topic. After removing duplicates, we were left with 1591 articles. Further screening based on titles and abstracts, according to our inclusion criteria, narrowed the selection down to 49 articles for in-depth review. We excluded articles which did not include risk assessment tools (N = 5), those not examining violence risk evaluation tools in forensic or acute psychiatric care settings (N = 1), and those that did not report the outcomes of these tools (N = 3). Ultimately, 40 articles met our inclusion criteria. We opted to exclude studies that solely focused on personality scales, as they did not assess violence risk directly. The PRISMA flowchart detailing the study inclusion process is available in Figure 1. Information on the identified studies and their specifics is outlined in

Table 1. Detailed results of scoping review study selection.

Authors	Psychiatric Context	Sample Size	Risk Assessment Tools	Number of Assessment Items Evaluated	Type of Reporting	Main Reported Effectiveness Metrics	Conclusion
(Bass et al., 2010) [22]	Forensic psychiatry	45 (39 men and 6 women)	PCL-R-2, VRAG	20 items (PCL-R-2) 12 items (VRAG)	Informant reported	Not reported	Demonstrated cognitive and motivational implication in impulsivity
(Blasioli et al., 2021) [23]	Forensic and acute psychiatry	337 (870 episodes)	RAI-MH, RHO CAP, SIRS	400 items (RAI-MH)	Informant reported	Reliability RHO CAP and the severity of the incidents of all patients: 31.26%, false positive: 57%; Prediction RHO CAP for patients representing no-to-low risk: 73.55%, false positive: 26.45%	RHO CAP is more effective in predicting patients with no or minimal risk of committing aggressive acts.
(Blasioli et al., 2021) [24]	Acute psychiatry	870 episodes of violence	RAI-MH	400 items (RAI-MH)	Informant reported	Accuracy: 75%, Sensibility: 28.57%, Specificity: 91.85%, Balanced Accuracy: 60.21, PPV: 56%, NPV: 77.99%	This study showed a model for predicting the risk of violence in patients with good accuracy, but the sensitivity needed improvement.
(Boccaccini et al., 2017) [25]	Forensic psychiatry	95 evaluators	PCL-R	20 items (PCL-R)	Informant reported	Not reported	PCL-R is used in various ways. Most assessors are more likely to use it to measure the risk of violent or sexual reoffending (77.9%), 45.3% use it as a measure of mental disorder, and about one-third (31.6%) use it for both risk assessment and mental disorder evaluation.
(Brassard et al., 2022) [26]	Forensic psychiatry	100 men	BIS-11,MOAS	30 items (BIS) 4 items (MOAS)	Self-reported, informant reported	Not reported	Demonstrated a link between violence in psychiatric units and brain damage, particularly in the frontal cortex. Therefore, it is important to conduct neuropsychological evaluations for patients at risk of violent behavior.
(Brathovde, 2021) [27]	Acute psychiatry	32	BVC	6 items (BVC)	Informant reported	Not reported	There was a 6.5% improvement in the rate of using restraints with the implementation of BVC. Moreover, 95% of the nursing staff have a better perspective on the use of this tool in this study.
(Burk et al., 2023) [28]	Acute psychiatry	377	BVC	6 items (BVC)	Informant reported	Not reported	The Birmingham Agitation Management protocol, including the BVC, helps to reduce the number of medications used in situations of agitation.
(Cabeldue et al., 2018) [29]	Forensic psychiatry	140 (114 men and 26 women)	HCR-20 V3, SOS	20 items (HCR-20) 20 items (SOS)	Informant reported	An LRA model assessed the prediction of violence in the hospital in the preceding review period. Overall, the model was significant (x2 = 98.28, p < 0.001; Nagelkerke R2 = 0.69).	HCR-20 enhances clinical decision making. It is useful in determining which patients should be transferred from a forensic psychiatry unit to a general psychiatry unit or even released into the community.
(Chagigiorgis et al., 2013) [15]	Forensic psychiatry	121 (115 men and 6 women)	BRC	41 items (BRC)	Informant reported	Significant correlations were found between the Harm to Others risk scale at the first assessment time and aggressive (r = 0.40, p < 0.01) and total (r = 0.40, p < 0.01) incidents.	BRC, especially the “Harm to Others” scale, allows for the assessment of dynamic changes in the short-term risk of physical and/or verbal aggression.
(Charrette et al., 2021) [30]	Forensic psychiatry	1800	HCR-20, VRAG	20 items (HCR-20) 12 items (VRAG)	Informant reported	Not reported	Observing a colleague using a risk management tool increases our likelihood of using one in our evaluations.
(Clarke et al., 2010) [31]	Acute psychiatry	48	BVC	6 items (BVC)	Informant reported	Scores on Day 1 were significantly related to scores on Day 2 (r = 0.515 p = 0.000), and scores on Day 2 were significantly correlated with scores on Day 3 (r = 0.475 p = 0.03), but Day 1 scores were not correlated with scores on Day 3.	BVC is easy and quick to use even in a very busy psychiatric unit. The tool provides day-to-day predictability with a general trend over the long term.
(Comai et al., 2021) [32]	Forensic psychiatry	124	MACVI, PCL-R	5 items (MACVI) 20 items (PCL-R)	Informant reported	Not reported	Cannabis use is common among patients with psychotic disorders, but it is not associated with a risk of violence and impulsivity. However, there is a risk of violence with alcohol consumption.
(Cook et al., 2018) [33]	Forensic psychiatry	39	HARM-FV, AIS, HCR-20 V3, MOAS	16 factors (HARM-FV) 9 items (AIS) 20 items (HCR-20 V3) 4 items (MOAS)	Informant reported	A substantial proportion of cases assigned to either the low or high SRRs on the HARM-FV received the corresponding rating on the HCR-20V3, 72% and 77%, respectively. When aggression data were dichotomized as present or absent, there was substantial agreement between the AIS and MOAS (κ = 0.79, p < 0.0001), in detecting any aggressive outcome in the follow-up period.	The validity of the HARM-FV and the AIS was demonstrated in this study. They were found to be comparable to the HCR-20 and MOAS.
(Côté et al., 2012) [34]	Forensic psychiatry	96	HCR-20	20 items (HCR-20)	Informant reported	Intraclass correlation coefficients yielded excellent interrater reliability for the HCR-20 total score (0.87). For individual HCR-20 items, interrater kappas ranged from 0.41 to 1.00 (mean 0.90); agreement was perfect for 12 of the 20 items.	Although the validity of the HCR-20 was not assessed, the results demonstrate the importance of using a violence management tool with structured clinical judgment in court settings.
(Delgado et al., 2020) [35]	Forensic psychiatry	200 (164 men, 3 women and 3 transgenders)	SAQ, MCAA	72 items (SAQ) 46 items (MCAA, part B)	Self-reported	The SAQ total score was significantly associated with violence (OR = 1.07, p = 0.004) and an area under the curb (AUC) of 0.71 (where an AUC of 0.50 would indicate a violence risk prediction no more accurate than luck).	Criminal risk rather than psychiatric symptoms was predictive of institutional violence within 6 months. It was advised against using only the total SAQ score for risk assessment; it would be better to also use another tool such as the Violence Risk Screening-10, HCR-20 V3, and START, along with objective data (the history of violence).
(Desmarais et al., 2010) [36]	Forensic psychiatry	137 (122 men and 15 women)	START, SOS	44 items (START)	Informant reported	Internal consistency of START: alpha = 0.87 Interrater reliability START between 3 professions: ICC2 = 0.87, p < 0.001; START total scores generally found to predict the outcome behaviors at rates greater than chance (physical aggression: r pb = 0.23, p < 0.001; AUC 0.65, IC (95%) = 0.57–0.72, p < 0.001); SOS: interrater reliability SOS: ICC2 = 0.70 p < 0.001; the model approached significance for the prediction of aggression toward others in the full sample (full: B = −0.58, p = 0.07; inpatient: B = −0.36, p = 0.49).	The study supported the structured and multidisciplinary approach to short-term risk assessment offered by the START tool.
(Desmarais et al., 2012) [37]	Forensic psychiatry	120	START, HCR-20, PCL-SV, OAS	44 items (START) 20 items (HCR-20) 20 items (PCL-SV) 4 items (OAS)	Informant reported	START strength and vulnerability total scores and violence risk estimates demonstrated good predictive validity for all outcome domains, with AUC values ranging from 0.75 (SE = 0.05) for Strength total scores predicting verbal aggression to 0.85 (SE = 0.04) for violence risk estimates predicting physical aggression toward others (all ps < 0.001).	START has shown predictive and incremental validity compared to the historical sub-scale scores of HCR-20 and the total scores of PCL-SV.
(Desmarais et al., 2012) [38]	Forensic psychiatry	96	START	20 items (START)	Informant reported	Internal consistency of the strength and vulnerability total scores were good, as indicated by Cronbach’s a coefficient of 0.95 and 0.90, respectively	This study showed the potential to use the START in prison dejudicialization programs.
(Dumais et al., 2012) [39]	Acute psychiatry	77	DASAfr	7 items (DASAfr)	Informant reported	Total score had little ability to predict aggression against objects but had acceptable predictive accuracy for predicting aggression against others and against nursing staff (0.70–0.79 = acceptable, 0.80–0.89 = excellent, and above 0.9 = outstanding).	The study demonstrated that the DASAfr had predictive accuracy comparable to the original English version. The nursing care team had positive opinions about the use of DASAfr, and the staff used it systematically.
(Dutch et al., 2019) [40]	Acute psychiatry	24 during interrater phase, 103 for validation	DASA-YV	11 items (DASA-YV)	Informant reported	Interrater reliability: κ = 0.79 at 95%, CI of 0.70 to 0.89, and a p value of <0.001 was significant. Predictive validity: The AUC for the DASA-YV as a measure of predictive validity started from 0.84.	The validity of the DASA-YV tool was assessed in this youth population under study, and it was found that it can be a useful tool for quickly assisting psychiatric nurses in predicting aggressiveness.
(Edens et al., 2010) [41]	Forensic psychiatry	Not specified	PPI-R, BIS, NAS-PI, PCL-R, VRAG, HCR-20	167 items version (PPI-R) 30 items (BIS) 85 items (NAS-PI) 20 items (PCL-R) 12 items (VRAG) 20 items (HCR-20)	Self-reported, informant reported	Not reported	The PPI-R is promising to assess the diverse personality traits historically linked to the concept of psychopathy.
(Grimbos et al., 2016) [42]	Forensic psychiatry	360 (292 men and 68 women)	HCR-20 V2	20 items (HCR-20 V2)	Informant reported	Interrater reliability: 0.88 (total score) Internal consistency: men (H: 0.70, C: 0.76, R: 0.42), Femmes (H: 0.83, C: 0.77, R: 0.58)	Clinician evaluations using HCR-20 V2 demonstrated good agreement with research data for both men and women. The item-level analysis supports the tool’s content validity and suggests that the items assessed capture multiple areas of risk relevant for women.
(Grossi et al., 2019) [43]	Forensic psychiatry	169 (140 men and 29 women)	HCR-20 V3, SOS	20 items (HCR-20 V3) 20 items (SOS)	Informant reported	Overall model was significant (F = 5.00, p = 0.001, adjusted R2 = 0.10).	HCR-20 V3 is useful for predicting victimization experiences.
(Guy et al., 2012) [44]	Forensic psychiatry	1	HCR-20, PCL-R	20 items (HCR-20, PCL-R)	Informant reported	Not reported	The SPJ approach is a scientifically grounded and empirically validated method for assessing the risk of violence. Moreover, including dynamic risk factors that target an intervention can help reduce or manage risk, making it a clinically useful paradigm.
(Hardin et al., 2022) [45]	Forensic psychiatry	136	NAS, SIV	48 items (NAS) 8 items (SIV)	Self-reported	The association between anger rumination and imagined violence was more robust (r = 0.35, p < 0.001). Anger rumination was positively and significantly related to SIV recency (r = 0.19), frequency (r = 0.24), similarity in type of harm (r = 0.14), focus (r = 0.14), escalation (r = 0.17), and proximity to target (r = 0.19). Anger rumination [b = 0.28, z (1118) = 6.42, p < 0.001] was also significantly related to prehospitalization violence. Anger rumination significantly predicted posthospitalization violence (b = 0.12, z (532) = 2.31, p = 0.02).	Dwelling on anger was predictive of violent behavior, both in looking back and in predicting future incidents, considering factors such as age, sex, race, physical violence in childhood, and a tendency towards anger/aggression.
(Healey et al., 2020) [46]	Forensic psychiatry	Study 1: 36 Study 2: 55	BRC, HARM-FV, eHARM-FV, AIS	41 items (BRC) 11 items (HARM-FV) 24 items (eHARM-FV)	Informant reported	Study 1: - Predictive validity BRC: 0.58–0.71 - Validity HARM-FV: 0.59–0.96 (AUC) Study 2: - Predictive validity BRC: 0.72–0.95 - Predictive validity eHARM-FV: 0.56–0.90 (AUC)	The BRC, HARM-FV, and eHARM-FV are brief and easy-to-use tools in forensic psychiatry, and studies show that they have good predictive validity.
(Hilton et al., 2016) [47]	Forensic psychiatry	63	VRAG, PCL-R, HCR-20	12 items (VRAG) 20 items (PCL-R) 20 items (HCR-20)	Informant reported	Disposition decisions VRAG: AUC 0.714, 95% CI (0.581, 0.847) Transfer decisions VRAG: AUC 0.708, 95% CI (0.573, 0.843)	The VRAG, PCL-R, and HCR-20 scores are related to the decisions made, including transfers and detention.
(Hogan et al., 2016) [48]	Forensic psychiatry	99	VRAG-R, PCL-R, HCR-20 V3, START, VRS, SOAS-R	12 items (VRAG-R) 20 items (PCL-R) 20 items (HCR-20 V3) 44 items (START) 26 items (VRS)	Informant reported	Predictive accuracies (AUC): VRAG-R: 0.60, 95% CI (0.44, 0.76) PCL-R: 0.63, 95% CI (0.49, 0.77) HCR-20 V3: 0.76, 95% CI (0.66, 0.86) START: vulnerability total 0.76, 95% CI (0.65, 0.87); strength total 0.69, 95% CI (0.57, 0.81); VRS: 0.68, 95% CI (0.44, 0.81)	The HCR-20 V3 and START had good predictive values. The PCL-R and VRAG-R did not have good predictive values compared to previous studies.
(Hogan et al., 2019) [49]	Forensic psychiatry	82	VRAG-R, PCL-R, HCR-20 V3, START, VRS, SOAS-R	12 items (VRAG-R) 20 items (PCL-R) 20 items (HCR-20 V3) 44 items (START) 26 items (VRS)	Informant reported	Pretreatment: - HCR-20 V3: violent 0.85 (0.76, 0.94), general 0.68 (0.53, 0.82) - START: (1) vulnerability total: violent 0.83 (0.72, 0.94), general 0.63 (0.49, 0.78), (2) strength total: violent 0.77 (0.63, 0.92), general 0.64 (0.48, 0.80) - VRS: violent 0.82 (0.67, 0.97), general 0.67 (0.50, 0.85) - VRAG-R: violent 0.74 (0.61, 0.87), general 0.70 (0.55, 0.85) - PCL-R: violent 0.74 (0.60, 0.87), general 0.69 (0.52, 0.85) Posttreatment: - HCR-20 V3: violent 0.81 (0.69, 0.92), general 0.65 (0.50, 0.81) - START: (1) vulnerability total: violent 0.75 (0.60, 0.91), general 0.71 (0.58, 0.85), (2) strength total: violent 0.67 (0.51, 0.83), general 0.63 (0.48, 0.78) - VRS: violent 0.82 (0.66, 0.97), general 0.67 (0.49, 0.85) - VRAG-R: N/A - PCL- R: N/A	The scales HCR-20 V3, VRAG-R, START, PCL-R, and VRS have good predictive values in assessing the risk of violence in forensic psychiatry.
(Jung et al., 2013) [50]	Forensic psychiatry	292	HCR-20, PAI, CLS, LSM	20 items (HCR-20) 344 items (PAI) N/A (CSL) 43 to 54 items (LSM)	Self-reported, informant reported	N/A	The risk management scale of the HCR-20 did not perform well.
(McDermott et al., 2011) [51]	Forensic psychiatry	146	COVR, VRAG, PCL-R, HCR-20	44 items (COVR) 12 items (VRAG) 20 items (PCL-R) 20 items (HCR-20)	Informant reported	Correlation between tool and aggression: - COVR: 0.331 - PCL-R: 0.290 - VRAG: 0.266 - HCR-20: 0.260 AUC: - COVR: 0.725 - PCL-R: 0.733 - VRAG: 0.720 - HCR-20: 0.728 p < 0.01	The COVR is a quick and useful tool for assessing the risk of violence.
(Nicholls et al., 2011) [52]	Forensic psychiatry	1057 forms	START	44 items (START)	Informant reported	Internal consistency for strength = 0.80 and for vulnerabilities = 0.76	The START tool aids in the assessment and management of risk in forensic psychiatry.
(Oziel et al., 2020) [53]	Forensic psychiatry	50	SAPROF, PCL-R, LS/CMI	17 items (SAPROF) 20 items (PCL-R) 53 items (LS/CMI)	Informant reported	Internal scale reliability SAPROF: Cronbach’s alpha = 0.72 Correlations SAPROF and START Strength scale = 0.84, START Vulnerability scale = −0.83, HCR-20 V3 = −0.52, LS/CMI = −0.14, PCL-R = −0.21 Prediction at 6-month follow-up: PRN administrations: HCR-20 V3 (AUC) 0.71 [0.57, 0.86], SAPROF (AUC) = 0.61 (0.45, 0.78), Institutional misconduct: HCR-20 V3 (AUC) 0.82 (0.65, 1.00), SAPROF (AUC) 0.87 (0.75, 0.99) Disposition breaches: HCR-20 V3 = 0.70 (0.54, 0.86), SAPROF total score (AUC) = 0.73 (0.58, 0.89), SE = 0.08	The SAPROF has good predictive value for administering PRNs over a six-month follow-up. It negatively correlates with the HCR-20 and START vulnerability scale, positively correlates with the total score and strength scale of START, and has no correlation with the LS/CMI and PCL-R.
(Penney et al., 2016) [54]	Forensic psychiatry	87	HCR-20 V3, PCL-R, BPRS, MACVI, SRD	20 items (HCR-20) 20 items (PCL-R) 24 items (BPRS) 22 items (MAC-VI) 15 items (SRD)	Self-reported, informant reported	HCR-20 is more efficient than PCL-R to predict violence. HCR-20 is the tool studied that has the best predictive validity. On the contrary, all self-reported tools were not able to predict violence (except the BPRS).	The usual risk factors associated with predicting violence and recidivism are useful for predicting the risk of rehospitalization. It is important to use tools that consider the evolution of dynamic risk factors over time, not just their presence.
(Penney et al., 2014) [55]	Forensic psychiatry	21 evaluators	HCR-20	20 items (HCR-20)	Informant reported	Interrater reliability: - Prior training: H scale 0.92, C scale: 0.92, R scale 0.62, total: 0.93, summary risk rating: 0.77 - No prior training: H scale 0.90, C scale 0.77, R scale 0.75, total 0.90, summary risk rating 0.90 - Expertise more than 10 years: H scale 0.92, C scale: 0.95, R scale 0.58, total 0.91, SRR: 0.73 - Expertise less than 10 years: H scale 0.94, C scale 0.80, R scale 0.68, total 0.93, SRR: 0.85	Less than 3% of the difference between the HCR-20 total score and the SRR is due to the evaluator’s impact. The assessments by various types of professionals, whether trained or not, are of good-to-excellent reliability. Most score differences are related to differences between patients.
(Sarver et al., 2019) [56]	Acute psychiatry	222	BVC, SOAS-R	6 items (BVC) N/A (SOAS-R)	Informant reported	Logistic Regression for Evidence of Violence: - Intercept: dF I, estimate −2.5201, standard error 0.2978, Wald chi-square test statistic 71.6056, p < 0.0001 - BVC score on admission: dF I, estimate −1.2260, standard error 0.2033, Wald chi-square test statistic 36.3521	For each point on the BVC tool, there is a 3.4 times greater chance of violence with a 95% confidence interval of 2.29–5.08.
(Starzomsky et al., 2015) [57]	Forensic psychiatry	121	IRRS, OAS, PBC	7 items (IRRS) 4 items (OAS) N/A (PBC)	Self-reported, informant reported	IRRS predicted physical aggression: AUC from 0.62 to 0.76 IRRS predicted verbal aggression: AUC from 0.62 to 0.80 Sensibility: 39.3% to 67.7% Specificity: 73.4% to 91.8% PPV: 43.3% to 75% NPV: 75.4% to 88.6%	The IRRS tool demonstrates good potential, comparable to the DASA, in predicting the risk of verbal and physical violence.
(Wilson et al., 2010) [58]	Forensic psychiatry	30	START, MOAS	44 items (START) 4 items (MOAS)	Informant reported	The results show that strength total scores and vulnerability total scores could predict aggression behavior for the next three months following assessment: strength, AUC = 0.73, 95% CI (0.63, 0.83); vulnerability, AUC = 0.74, 95% CI (0.64, 0.85); and both p < 0.001. It was the same for global risk estimates, AUC = 0.82, IC 95% CI (0.73, 0.91), and p < 0.001.	The total strength and vulnerability scores of the START, as well as the total score, showed excellent interrater reliability. Additionally, the total strength score and total vulnerability score yielded better results for short-term follow-ups (9 months for strength and 6 months for vulnerability) compared to long-term follow-ups (over 12 months).
(Winters et al., 2023) [59]	Forensic psychiatry	109	IRRS, MOAS	7 items (IRRS) 4 items (MOAS)	Informant reported	Verbal aggression: - RR: 5.9 to 8.4 - AUC: 0.84 to 0.87 - Sensibility: 60 to 75% - Specificity: 83.3 to 88.8% - PPV: 54.5 to 60% - NPV: 78.9 to 92.6% Physical aggression: - RR: 6.9 to 8.7 - AUC: 0.82 to 0.85 - Sensibility: 63.4 to 75% - Specificity: 77.1 to 84.7% - PPV: 27.3 to 31.8% - NPV: 95.4 to 96.6%	The study shows that the IRRS tool predicted physical and verbal violence up to 2 weeks after its use.
(Woods et al., 2015) [60]	Forensic psychiatry	46	BVC, SOAS-R	6 items (BVC) NA (SOAS-R)	Informant reported	BVC: Wald (1) = 29.57, p < 0.001, eB = 2.68, 95% CI (1.88 to 3.82) (predicted violence) Slide rule did not predict this outcome: Wald (1) = 0.07, p = 0.793, eB = 0.969, 95% CI (0.77 to 1.22) after controlling for the BVC. BVC (AUC): 0.73 (0.64–0.82)	The BVC tool showed high predictive accuracy for aggression among hospitalized patients, far exceeding chance. In this study, the BVC was a better predictor of any aggressive incident involving hospitalized patients rather than the relative severity of the incident.

Abbreviations: AIS: Aggressive Incidents Scale; AUC: area under the curve; BIS-11: Barratt Impulsiveness Scale; BRC: Brockville Risk Checklist; Brief Psychiatric Rating Scale: BPRS; BVC: Brøset Violence Checklist; CLS: Cormier Lang Score; COVR: Classification of Violence Risk; DASAfr: Dynamic Appraisal of Situational Aggression-French version; DASA-YV: Dynamic Appraisal of Situational Aggression-Youth Version; HARM-FV: Hamilton Anatomy Risk Management Forensic Version; eHARM-FV: Electronic Hamilton Anatomy of Risk Management-Forensic Version; HCR-20: Historical Clinical Risk Management 20; HCR-20 V2: Historical Clinical Risk Management 20 Version 2; HCR-20 V3: Historical Clinical Risk Management 20 Version 3; IRRS: Imminent Risk Rating Scale; LRA: logistic regression analysis; LS/CMI: Level of Service/Case Management Inventory; LSM: Level of Service Measure; MACVI: MacArthur Violence Risk Assessment Scale; MCAA: Measure of Criminal Attitudes and Associates; MOAS: Modified Overt Aggression Scale; NAS: Novaco Anger Scale; NAS-PI: Novaco Anger Scale and Provocation Index; NPV: negative predictive value; OAS: Overt Aggression Scale; PAI: Personality Assessment Inventory; PBC: Problem Behavior Checklist; PCL-R: Psychopathy Checklist-Revised; PCL-R-2: Psychopathy Checklist-Revised 2nd edition; PCL-SV: Hare Psychopathy Checklist: Screening Version; PPI-R: Psychopathic Personality Inventory-Revised; PPV: positive predictive value; RAI-MH: Resident Assessment Instrument for Mental Health; RHO CAP: Risk of Harm to Others Clinical Assessment Protocol; SAPROF: Structured Assessment of Protective Factors for Violence Risk; SAQ: Self-Appraisal Questionnaire; SRD: Self-Report of Delinquency; SIV: schedule of imagined violence; SOS: Start Outcome Scale; SOAS-R: Staff Observation Aggression Scale-Revised; SPJ: Structured professional judgment; START: Short-Term Assessment of Risk and Treatability; VRAG: Violence Risk Appraisal Guide; VRAG-R: Violence Risk Appraisal Guide Revised; VRS: violence risk scale.

.

3.2. Violence Risk Evaluation Tools

3.2.1. Hamilton Anatomy of Risk Management Forensic Version (HARM-FV) and Electronic Hamilton Anatomy of Risk Management Forensic Version (eHARM-FV)

The HARM-FV is a structured clinical judgment tool that encompasses 6 historical and 10 dynamic risk factors [33]. This tool assists clinicians in assessing the risk of aggression and in formulating strategies for managing this risk in the ensuing days and weeks [33]. Within this framework, the Aggressive Incidents Scale (AIS) is integrated; it is a nine-point scale that quantifies aggression levels [46]. The eHARM-FV enables the monitoring of a patient’s progress and the comparison of groups through visual representations generated from data inputted into an electronic version of the HARM-FV. A study compared the HARM-FV with the Brockville Risk Checklist in two samples: one with 36 patients and another with 55 patients in a forensic psychiatry unit in Ontario that assessed the HARM’s predictive validity using the area under the curve (AUC), which ranged from 0.56 to 0.96 based on whether the context was immediate vs. short-term and whether support was available [46]. For the eHARM-FV, the same research found that its predictive validity fluctuated between 0.56 and 0.90 and was again influenced by the duration of the assessment period and the availability of support [46].

3.2.2. Historical Clinical Risk Management 20 (HCR-20)

The HCR-20 emerged as the most frequently examined risk assessment tool across 19 studies. This tool is a structured clinical judgment instrument consisting of 20 items, incorporating both dynamic factors (5 clinical symptoms and 5 risk management items) and static factors (10 historical factors). Its predictive utility for violence was demonstrated as significant (χ² = 98.28, p < 0.001; Nagelkerke R² = 0.69) in a study, which analyzed 140 patients in a forensic psychiatry unit in New York who were found not criminally responsible due to mental disorders [29]. Another study in a forensic psychiatry unit with 30 patients in British Columbia showed high interrater reliability for both the total score (0.88) and the final judgment (0.84) [58]. Hogan and colleagues assessed 99 patients in a forensic psychiatric unit in Western Canada, and reported a predictive accuracy of 0.76 with a 95% confidence interval (CI) of 0.66 to 0.86, and detailed the reliability of different HCR-20 categories: historical scale (0.93), clinical scale (0.88), and risk scale (0.64) [49]. Also, a comparative study of 68 women to 292 men in a forensic psychiatric unit in Ontario, using the HCR-20, found internal consistencies for historical, clinical, and risk items to be 0.83, 0.77, and 0.58, respectively, as opposed to 0.70, 0.76, and 0.42 in men [42].

3.2.3. Psychopathy Checklist-Revised (PCL-R)

The PCL-R is a 20-item tool designed to assess the personality traits associated with psychopathy [49]. It was the second most frequently discussed risk assessment instrument, cited in 14 studies. The tool has been adapted into a Youth Version and a Screening Version to accommodate different populations. In the previously mentioned study by Hogan and Olver, the PCL-R demonstrated a predictive accuracy with a 95% CI of 0.63 (0.49, 0.77). A meta-analysis by Ramesh T. et al. (2018) reported the sensitivity of the PCL-R to be 0.53 (95% CI 0.45, 0.63), specificity to be 0.60 (95% CI 0.38, 0.81), PPV to be 0.72 (95% CI 0.63, 0.82), and NPV to be 0.55 (95% CI 0.43, 0.70) [49].

3.2.4. Short-Term Assessment of Risk and Treatability (START)

The START is a structured clinical judgment tool consisting of 22 dynamic items and is evaluated across two dimensions: strengths and vulnerabilities. A meta-analysis reported the START’s sensitivity at 0.81 (95% CI 0.78–0.89), specificity at 0.60 (95% CI 0.55–0.68), PPV at 0.52 (95% CI 0.44–0.62), and NPV at 0.86 (95% CI 0.82–0.92). Another study involving patients in a forensic psychiatry unit in Western Canada assessed the predictive accuracy of these two factors, finding a score of 0.76 (95% CI 0.65, 0.87) for the vulnerabilities total and 0.69 (95% CI 0.57, 0.81) for the strength total [49]. In evaluating the concordance between the START and HCR-20 among a sample of 120 male patients in a Western Canadian forensic psychiatry unit, the agreement on violence prediction was κ = 0.77 with p < 0.001 [37]. The interrater reliability for the strength factor, the vulnerability factor, and the estimated violence risk were 0.93, 0.95, and 0.85, respectively, using the Intraclass Correlation Coefficient (ICC2) [37].

3.2.5. Brøset Violence Checklist (BVC)

The BVC is a brief tool for predicting short-term violence, comprising six items, i.e., confusion, irritability, boisterousness, verbal threats, physical threats, and attacks on objects, which are scored as either present or absent. Its effectiveness has been assessed in six of the forty identified articles. A meta-analysis highlighted the BVC’s sensitivity at 0.66 (95% CI 0.41–0.80), specificity at 1.00 (95% CI 0.76–1.00), PPV at 0.37 (95% CI 0.25–0.68), and NPV at 0.99 (95% CI 0.78–0.99) and a diagnostic odds ratio (DOR) of 43.1 (95% CI 29.6–56.6) [61]. In a study involving 222 patients in an acute psychiatric unit in the United States, Sarver et al. (2019) reported the tool’s effect size (dF I) at −1.2260, with a standard error of 0.2033 and a significant Wald chi-square statistic of 36.3521. Meanwhile, Woods and colleagues reported, in their research with 46 patients in a forensic psychiatry unit in Saskatchewan, the BVC to have a Wald score of 29.57 (p < 0.001, Exp(B) = 2.68, 95% CI = 1.88 to 3.82) and an area under the curve (AUC) of 0.73, indicating its utility in assessing the risk of violence [60].

3.2.6. Dynamic Appraisal of Situational Aggression (DASA)

The DASA is an actuarial tool designed to evaluate the risk of imminent aggression within the next 24 h among mental health patients. It includes seven factors: irritability, impulsivity, unwillingness to follow directions, sensitivity to perceived provocation, being easily angered when requests are denied, negative attitudes, and verbal threats. This instrument was mentioned in three articles, including its French version (DASAfr), its Youth Version (DASA-YV), and the standard version. Dumais and colleagues discussed the French version in their study of 77 patients in intensive psychiatric units in Quebec [39]. Their findings indicate the instrument’s moderate predictive ability for aggression towards objects (a next shift AUC of 0.66 (0.58–0.73) vs. a next 24 h AUC 0.66 (0.61–0.71)), and acceptable accuracy for aggression towards others (next shift 0.73 (0.57–0.89) vs. next 24 h 0.71 (0.61–0.81)) and nursing staff (next shift 0.72 (0.59–0.85) vs. next 24 h 0.70 (0.61–0.78)). They concluded that the French version’s predictive accuracy was comparable to the original tool. The DASA-YV adds four items: anxiety or fearfulness, low empathy/remorse, significant peer rejection, and external stressors. Dutch and Patil observed that the DASA-YV’s predictive validity varied from 0.84 (physical aggression towards others) to 0.92 (verbal aggression towards others) in a study of 127 patients in a pediatric and adolescent acute psychiatric unit in the United States [40]. Lastly, a meta-analysis showed that the standard DASA had a sensitivity of 0.73 (0.43–0.75), a specificity of 0.75 (0.69–0.76), a PPV of 0.35 (0.20–0.63), an NPV of 0.99 (0.97–1.00), and a DOR of 10.9 (9.8–12.1) [61].

3.2.7. Other Instruments

The Structured Assessment of Protective Factors for Violence Risk (SAPROF) was examined by Oziel and colleagues and found to have internal scale reliability, as measured by Cronbach’s alpha of 0.72 [53]. This study involved 50 patients in a forensic psychiatry unit in Ontario. In another study, the Imminent Risk Rating Scale (IRRS) was evaluated by Winters and colleagues on 109 patients in a forensic psychiatry unit in the United States [59]. It demonstrated the capability to predict risk for up to two weeks, with sensitivity ranging from 60 to 75% for verbal aggression and 63.4 to 75% for physical aggression and specificity from 83.3 to 88.8% for verbal aggression and 77.1 to 84.7% for physical aggression. This tool was assessed with 301 patients across four locked psychiatric units in the United States. Lastly, the Violence Risk Appraisal Guide Revised (VRAG-R) reported an accuracy of 0.60 (95% CI = 0.44, 0.76) and an AUC of 0.720 [49,51].

3.3. Outcomes

Numerous studies have investigated the effectiveness of various risk assessment tools. Wilson and colleagues reported that the dynamic factors of items C and R in the HCR–20, as well as the START, demonstrated a solid predictive value for future institutional risk, with no significant differences in predictive value between the START and HCR-20 identified [58]. Also, Penney and colleagues concluded in their research that the HCR-20 was more effective in predicting violence than the PCL-R [54]. For forensic patients, certain tools, including the HCR-20, PCL-R, START, VRAG-R, and VRS, proved to be efficient in assessing the risk of violence recidivism [49].

4. Discussion

This scoping review examined the utilization of risk management tools in forensic psychiatry and acute psychiatric units in North America, with the goal of determining the predominant tools used and the metrics associated with their usage. The commonly employed risk assessment scales were the HARM-FV, eHARM-FV, HCR-20, PCL-R, START, BVC, and DASA. Validity and reliability metrics as well as the ability to predict violent events varied greatly across the different tools depending on their context of use and the specific population evaluated.

The identified risk assessment tools in forensic and acute psychiatric settings for the North American context corresponds to what has been previously identified in a systematic literature review on the subject for a broader context [62]. These tools are important as violence remains a major concern in psychiatric units, and while there is a wide variety of scales available for clinicians and staff, major challenges can impact their validity. When developing risk assessment tools, recent research stressed the importance of making them psychometrically sound, concise, consensus-rated, time-efficient, and practical for planning risk management, as well as incorporating user feedback to sustain implementation [63]. This contrasts well with forensic psychology as understanding and forecasting human behavior using psychological categories including personality traits, thought patterns, and past behavior is a common focus of forensic psychology. Forensic psychology also relies heavily on instruments such as the PCL-R, especially when evaluating long-term risk factors [62].

As it was identified in our literature review, many of the studies on risk assessment tools reported favorable accuracies in predicting the risk of violence, and these findings are correlated with the literature that is specifically designed for evaluating the effectiveness of targeted risk assessment tools. As an example, a meta-analysis evaluated the sensitivity, specificity, PPV, and NPV of the HCR-20 as 0.78 (95% CI 0.56 to 1.00), 0.71 (95% CI 0.56 to 1.00), 0.31 (95% CI 0.26 to 0.56), and 0.94 (95% CI 0.75 to 1.00), respectively [61]. However, these examples contrast the evidence pooled when assessing the reliability and predictive power of these tools. As an example, despite encouraging findings for the accuracy in predicting the risk of violence of risk assessment tools, the recent literature supports that violence risk assessment tools in forensic mental health have mixed evidence of predictive performance, and further research should be conducted in this area [64]. These contrasting results hint towards the fact that more studies are necessary to implement more robust risk assessments tools. It is important to emphasize that tools designed for predicting the risk of violence in penitentiary settings, which focus on measuring societal risk, may not necessarily be suitable for use in closed environments such as hospitals. The ecological validity of these tools must be critically evaluated and contrasted to ensure their applicability across different contexts.

Limitations

This scoping review specifically targets the various risk assessment tools employed in forensic and acute psychiatry settings in North America. The search strategies employed are tailored to this specific context; however, it is worth noting that certain clinical facilities may utilize different tools, which would not be captured in the search if they were not published. Additionally, due to the diversity of population samples across the studies identified, it was not feasible to make direct comparisons of tool performances to form a generalized assessment of their effectiveness.

5. Conclusions

As observed globally, violent incidents occur in North American forensic and acute psychiatric units. Given the higher occurrences of violence in psychiatric settings, utilizing risk assessment tools is crucial to mitigate harm for both staff and patients in these units. This scoping review aims to identify the risk assessment tools utilized for violence in North American psychiatric units. The search strategy led to the identification of several tools, with HARM-FV, eHARM-FV, HCR-20, PCL-R, START, BVC, and DASA being the most cited. Various performance metrics were reported for these tools, indicating variability in their accuracy for predicting violence risk within their specific contexts of use. With so many tools available, it makes us wonder about the limitations of the systematic use of risk assessment tools in psychiatric units. Furthermore, this study lays the groundwork for future research into developing tools for predicting violence risk in North American contexts, where violence remains prevalent and challenging to anticipate.