A review of progress in violence risk assessment methods

Abstract

Violence often occurs among patients with mental disorders. The risk of violence is assessed under the demand of psychiatric clinical treatment or forensic assessment. Corresponding therapeutic intervention strategies could be developed according to the outcome of the assessment. Currently, violence risk assessment methods are mainly divided into actuarial assessment and structured professional judgement. Scientific and objective assessment results support judicial decisions and risk management. However, all the assessment methods have certain shortcomings, and there is still room for improvement. This paper reviews several tools of violence risk assessment and their research progress, focusing on the main content of each tool and its applicability. The review aimed to provide a reference for the selection and application of violence risk assessment tools and optimization of violence risk assessment methods in the future.

Key points  

• The violence risk assessment methods are widely used in psychiatric clinical treatment or forensic assessment, but all the methods have certain shortcomings.

• The review discussed main contents and research progress of several violence risk assessment tools in order to explore applicability of each tool.

• The review aimed to provide a reference for the selection and application of violence risk assessment tools and optimization of violence risk assessment methods in the future.

Introduction

Violence can be defined as a subjective threat and/or harm caused to oneself, others, or society, with an actual or high probability of severe consequences, such as injury, death, psychological harm, growth impairment, or deprivation of human rights. Individuals with mental disorders are more likely to engage in violence, including short-term, long-term, and attempted violence. Violence risk assessment is a process performed by qualified professionals to assess the risk of violence, offer clinical treatment protocols, and, if possible, dynamically assess the efficacy, predict the possibility of future violence, and assess the risk of recidivism [1]. Meanwhile, scientific support for the formulation of risk management plans could be provided based on the nature, severity, and possibility of recidivism of the violence. A systematic risk assessment requires an individual risk model that considers risk behaviours, subjective thoughts, relevant risk factors, personal status, and protective factors [2]. The history of violence, recent occurrences of risk behaviours, and key factors such as severity, frequency, and behavioural patterns of risk behaviours should also be considered to provide scientific support.

In many countries, violence risk assessment is used in various contexts, such as civil involuntary dispositions, criminal sentencing, and determinations of mandatory hospitalization or discharge from mandatory treatment. In the legal field, these methods supplement and enhance judges’ weighing and assessment of other sentencing evidence, while in the clinical field, the methods are mostly used to determine whether interventions are necessary, and risk management may also be involved if targeted at specific individuals. A mature system of violence risk assessment has been established abroad, improving the predictive validity of risk by completing a shift from unstructured to structured assessment, with the former mainly relying on clinical experience and intuition to determine risk levels, while the latter is fulfilled based on objective factors [3].

In China, violence risk assessment is primarily associated with psychiatric clinical treatment and forensic assessment. In the clinical field, methods are similar to those used abroad and are employed to screen high-risk patients and assess whether patients can be discharged from mandatory hospitalization, which contributes to the prevention and reduction of potential violence. However, in forensic assessment, the focus is on the responsibility capacity and compulsory medical treatment due to the differences in domestic and external legal provisions. The deficiencies of violence risk assessment in China occur due to the lack of development of assessment methods compared to those in foreign counties. Additionally, it is not easy to conduct multi-centre studies. This paper reviews the applicability of frequently used violence risk assessment methods and their research progress to provide a reference for the optimization of violence risk assessment methods in China in the future.

Search strategy and study selection

Five electronic databases (PubMed, EMBASE, Web of Science, CNKI, and Wanfang databases) were used to identify articles. Extensive search terms were identified and used to reduce the risk of missing relevant studies. Search terms included (“PCL-R” OR “Psychopathy Checklist-Revised” OR “VRAG” OR “Violence Risk Appraisal Guide” OR “C-RASSPPC” OR “Chinese Risk Assessment Scale of Severe Psychiatric in Community” OR “HCR-20” OR “Historical, Clinical, Risk Management-20” OR “V-RISK-10” OR “Violence Risk Screening-10” OR “SORAG” OR “Sex Offender Risk Appraisal Guide” OR “VRS” OR “Violence Risk Scale” OR “SAVRY” OR “Structured Assessment of Violence Risk Among Youth” OR “violence risk assessment”) AND (“psychiatr*” OR “mental” OR “psycholog*”). The references of retrieved articles were also searched by hand for additional studies.

Based on the literature research, violence risk assessment methods were selected and enrolled as the following criteria: (1) had been widely used in the assessment of mental disorders; (2) complied by Chinese researchers or had been revised according to cultural customs as a Chinese version; (3) include abroad assessment methods to compare; (4) showed high reliability and validity, which meet the standards of relevant psychological and practical needs; (5) possess good operability; (6) the application value had been well recognized by the majority researchers.

Assessment methods which have a certain degree of duplicity had been selected. For example, the Psychopathy Checklist-Revised (PCL-R) was revised into three other structured professional judgement (SPJ) assessment methods: Historical, Clinical, Risk Management-20 (HCR-20), Sexual Violence Risk-20 (SVR-20), and Risk for Sexual Violence Protocol (RSVP). Therefore, this study focused on introducing PCL-R and HCR-20, while SVR-20 and RSVP were not included. Meanwhile, assessment methods which have a limited amount of literature, only used in certain organizations, or lack sufficient predictive validity were excluded in this study. For example, the Revised Level of Service Inventory (LSI-R) was excluded according to the weak predictive validity [4].

Violence risk assessment methods

Currently, violence risk assessments are primarily conducted by using structured methods, which can be divided into two categories: actuarial risk assessment and SPJ.

Actuarial risk assessment

Actuarial risk assessment is a process of scoring specific risk factors associated with delinquent behaviours and integrating these figures into a total score using a statistical algorithm, reflecting the probability of committing a violent crime. Actuarial risk assessment instruments (ARAIs) are constructed based on known data on risk factors for recidivism, non-recidivism, and patients with mental disorders, using precise algorithms to estimate the probability of future violent behaviour of the subject.

Several studies have suggested that using specific figures to present assessment results is a better decision-making method than other forms of risk assessment, and the use of numerical assessment reduces the subjective aspect due to different assessors, thus dissolving the problem of inconsistent risk classification [5]. Meanwhile, comparisons between patients can be made directly, if necessary, regardless of different assessors, since each patient is assessed by the same criteria using the same actuarial risk assessment method [6]. A large number of research studies have also revealed that ARAIs are superior to clinical methods for risk assessment [7–9].

However, ARAIs are limited in that they may ignore protective factors and other aspects of risk, such as the nature, severity, and urgency of the violence and the persistence and frequency of future violence [10]. Furthermore, the accuracy of ARAIs is debatable due to the lack of an evidence base for risk factors, sample size restrictions, and inadequate inclusion of risk factors [11]. Moreover, the predictive optimization of ARAIs is mostly performed in the same sample in which the tool was developed, possibly leading to changing outcomes when ARAIs are cross-validated in new samples [12]. Many experimental items on potential risk factors are not included in the rating items of ARAIs, yet may be of assistance for future validity enhancements [13]. Future research on actuarial risk assessment could aim to increase the number of assessing items, particularly dynamic factor items, or use various tools to improve the accuracy of risk rating assessments [14].

Recently, some ARAIs have incorporated “professional overrides” to improve predictive validity, allowing assessors to adjust risk classifications through subjective judgement, thus providing more flexibility in measuring risk levels and targeting individual subject characteristics [15].

Commonly used ARAIs include the PCL-R, Violence Risk Appraisal Guide (VRAG), and Sex Offender Risk Appraisal Guide (SORAG), among others.

SPJ

SPJ is a process combining unstructured clinical judgement with algorithms. It considers risk assessment as a dynamic assessment and allows assessors to apply information from the literature in conjunction with scales, provided that evidence is sufficient [16]. The aim of SPJ is not to predict violence but to prevent future violence by identifying the presence of relevant risk factors, assessing the individual’s current risk level and formulating the most appropriate interventions [17]. In a structured professional assessment, total scores are not used to assess the risk. Instead, evaluators ultimately decide on a low, medium, or high future risk rating after combining information from specific cases.

Compared to ARAIs, SPJ methods offer more flexibility for assessors and avoid bias resulting from ignoring the individual characteristics of the subject. The main advantage of SPJ methods is their attention to individual assessment and consideration of risk-related information not included in the assessment items, as well as the inclusion of dynamic factors, which can help implement intervention plans for subjects, observe dynamic changes in risk level, and develop corresponding risk management strategies when necessary. However, allowing clinical discretion might make SPJ methods more susceptible to potential rating bias, as assessors might have unique cognitive differences that lead to different opinions [18]. Especially in the case of non-standardized and highly subjective assessment methods, the bias caused by the differences in individual perceptions, professional and technical levels, and personal evaluation styles of different assessors is undoubtedly more significant [19]. However, the role of the evaluator or the practical value of subjective assessment methods should not be completely denied just because of the existence of bias [20]. Existing studies have also proposed strategies to address the bias problems (including but not limited to assessor-induced bias) in the field of court science or psychological assessment, such as Linear Sequential Unmasking-Expanded (LSU-E), which suggests that assessors should review the relevant information after completing the examination of the subject in order to reduce the influence of external information on the assessor [21]. Hierarchy of Expert Performance addresses the issue of reliability and bias of different assessors in examinations and conclusions, etc., which is an important reference to reduce the bias among different assessors in SPJ [22].

In addition to the potential for inter-assessor differences, the main drawback of SPJ methods is the lack of guidelines for integrating all evaluation items into an overall risk assessment, mechanisms for evaluating dynamic item changes assessment, and specific guidance on how to translate such changes into final risk judgements. The use of SPJ methods requires separate case interviews and information collection, which can be time consuming and labour intensive in practical applications [23].

Therefore, different methods have their own advantages and disadvantages. Scholars who advocate for SPJ believe that risk can rapidly change based on individual circumstances and should not be assessed as a static or one-time assessment but rather should focus on dynamic processes. Scholars who advocate for ARAIs believe that the inclusion of discretion can reduce the reliability and effectiveness of evaluations, and the objectivity of risk assessments under this method remains to be discussed. However, many data analysis studies have shown that the predictive validities of SPJ tools and ARAIs are actually similar [24, 25].

In recent years, attention to the risk formula in SPJ research has significantly increased, mainly to explain the potential mechanisms for individual risk generation and propose hypotheses for changes in risk, focusing on various factors playing different roles in the occurrence of violence [12].

Commonly used SPJ tools include the Brief Spousal Assault Form for the Evaluation of Risk (B-SAFER), Child Abuse Risk Evaluation-Version 2 (CARE-V2), Early Assessment Risk Lists for Boys and Girls (EARL-20B/EARL-21G), HCR-20, Estimate of Risk of Adolescent Sexual Offence Recidivism (ERASOR), and Structured Assessment of Violence Risk among Youth (SAVRY), among others.

Common violence risk assessment tools and their application

Violence risk assessment tools are developed for specific idiosyncratic groups, considering factors such as age, gender, behavioural health status, and environment, to provide information on the predictability of violence risk or to address risk needs. These tools are applied in practice after accuracy and validity have been verified. Some components of violence risk assessment tools can be used as part of a risk screening tool to make a simple and rapid decision on whether immediate risk intervention is necessary for the subject. However, as risk assessment tools primarily provide predictive rather than diagnostic information, it has been argued that statistics based on diagnostic classifications are not suitable for evaluating the accuracy of specific risk assessment tools [26]. To date, no single tool has been shown to have a superior ability for criminal behaviour [27].

Scholars have classified the development process of violence risk assessment into four generations. The first generation is unstructured clinical assessment. The second generation mainly contains actuarial methods and assesses through static factors. The third generation introduces dynamic factors, and the fourth generation contains both dynamic and static factors to assess specific types of violence [28]. In China, some scholars have also classified the development process of violence risk assessment into either four generations similar to those abroad or three generations comprising unstructured clinical assessment, actuarial assessment, and structured clinical assessment. This paper adopts the four-generation classification approach and introduces commonly used violence risk assessment methods accordingly.

First-generation unstructured clinical assessment

The first generation of violence risk assessment involves unstructured clinical judgements that heavily rely on the assessor’s knowledge, background, and experience. It primarily utilizes medical history information, has limited predictive validity, lacks specific assessment tools, and exhibits low accuracy and reliability. In China, the three-level risk category of clinical management of patients with severe mental impairment is mostly based on empirical clinical judgement, which belongs to the first generation of violence risk assessment.

Second-generation static assessment stage

The second generation is represented by structured clinical assessment, which utilizes actuarial methods, specifically static scales. As previously mentioned, actuarial assessment results remain consistent across assessors and can exclude individual differences between assessors, thereby enhancing accuracy compared to clinical assessments. Well-known examples of actuarial methods include PCL-R, VRAG, and Modified Overt Aggression Scales (MOAS), all of which have demonstrated good predictive validity. However, actuarial methods do not assess changes in risk and may not include all highly correlated factors with the risk of violence. Moreover, these methods are typically developed and optimized on specific source samples, which might compromise reliability when used to evaluate violence risk in other samples.

Psychopathy Checklist-Revised

PCL-R is a 20-item rating scale used to assess psychopathy, with assessors scoring items on a three-level scale from 0 to 2. A total score of 30 is commonly used as a threshold for differentiating between psychosis and non-psychosis [29]. Initially, the PCL scale was used only to rate psychosis. However, recent studies have shown a correlation between violence and psychopathy [30]. Thus, due to the correlation between its total score and antisocial behaviour, the PCL was later revised to be used in two actuarial assessment tools (VRAG and SORAG) and three SPJ tools (HCR-20, SVR-20, and RSVP) for the assessment of violence risk.

Screening and/or modification of the items in the PCL-R led to the development of two additional tools: the Psychopathy Checklist: Screening Version (PCL:SV) and the Psychopathy Checklist: Youth Version (PCL:YV). The PCL:SV comprises 12 items screened from the PCL-R, and the total score of this screening tool is highly correlated with the total score of the PCL-R [31]. The PCL:YV is a risk assessment tool designed to assess interpersonal, emotional, antisocial, and behavioural characteristics of adolescents aged 12–18 years. It was developed by modifying items such as short-term marital relationships. The PCL-R emphasizes not only antisocial behaviour but also interpersonal and emotional characteristics. However, there is considerable debate among scholars about the extent to which these factors reflect core psychopathic characteristics or what exactly core psychopathic characteristics are [32]. Despite yielding mixed results in meta-analyses on violent recidivism, numerous studies have shown that descriptive features and correlates of the PCL-R are similar and have good reliability among male and female violent offenders [33, 34].

The PCL-R was introduced in China and revised by Liu Banghui and Huang Xiting in 2009 [35]. They selected 30 male offenders in prison to test the reliability and validity of the scale. The results indicated good reliability for assessing psychopathy, but the scores were strongly influenced by age, education, marital status, and the number of offences.

VRAG

The VRAG is an actuarial tool consisting of 12 items used to assess the probability of violent and sexual offence recidivism in males. Scores for each category are expressed as percentages ranging from 0 to 100%, based on the positioning of a large sample of violent individuals [36]. Scholars conducted a 49-year follow-up study and developed a revised version of the VRAG, known as the VRAG-R, to improve the validity of assessing sexual and non-sexual offenders and simplify the scoring process. The VRAG-R does not require assessors to have knowledge of clinical diagnostic criteria when making scores. Scores for the VRAG-R are now more likely to be assigned on a scaled basis [37].

Rice et al. [38] conducted a meta-analysis of 52 studies that utilized VRAG or SORAG to investigate the accuracy of VRAG assessment, and the area under the curve (AUC) ranged between 0.70 and 0.72 depending on the assessors and expected application compliance [39]. Under ideal conditions of high reliability, no substitution or modification of items, and a fixed follow-up period, the AUC is ~0.85 [40]. The observed and expected rates of violent recidivism are well calibrated.

For the predictive validity of the VRAG-R in recent years, data show that the AUC values for general recidivism and violent recidivism are distributed between 0.65 and 0.66, 0.70 and 0.78, and 0.75 and 0.78 in different regions, while the AUC values for sexual recidivism are distributed between 0.60 and 0.67, and 0.61 and 0.63 [41–43]. Although the predictive validity of the VRAG-R is reported as comparable to those of the VRAG and SORAG, the VRAG-R is easier to score and does not require the administration of the entire PCL-R.

The Chinese Risk Assessment Scale of Severe Psychiatric in Community (C-RASSPPC)

C-RASSPPC is a violence risk assessment tool developed in China in 2010 by Li et al. [44]. The C-RASSPPC consists of eight static risk factor items, including the history of perpetration, abnormal behaviour in the last month, emotional and affective behaviour in the last month, hallucinations and delusions in the last month, treatment history, supervision, substance abuse history, and stressful life events. The history of perpetration item includes three subentries that assess the patient’s history of violence based on the type of event, time of occurrence, and frequency. The C-RASSPPC is used to evaluate the patient’s treatment history, history of illicit drug use, recent mood, and recent psychotic symptoms.

Third-generation static/dynamic assessment stage

Both second- and third-generation violence risk assessments incorporate algorithms to evaluate static risk factors, but the third-generation assessment introduces dynamic factors related to violent risk behaviours. It utilizes the SPJ method that considers individual characteristics, making it a more personalized approach to violence risk assessment. Utilizing such tools to assess the risk of violence can be beneficial for effective intervention and treatment management of violence.

HCR-20

The HCR-20 is an SPJ tool designed for constructing a comprehensive violence risk assessment, requiring corresponding empirical knowledge on the part of the assessor. Static historical factors and dynamic clinical manifestations were included, which are mainly used to assess the violence risk of mental disorder patients in general medical institutions, forensic evaluations, and custodial institutions [45–47]. The third version, HCR-20 V3, was revised from the original version and includes 20 violence risk factors, with 10 historical items (H), five clinical items (C), and five risk management items (R) [48]. Each item is scored on a three-level scale from 0 to 2. Assessors are required to comprehensively gather information, identify risk factors, sort out the causes of the subject’s past violence, predict the potential for future violence, and make effective recommendations for risk management strategies. The risk rating is summarized as low, medium, or high risk, although this rating is likely to change over time or with treatment.

The reliability and predictive validity of the HCR-20 V3 were validated in over 600 participants in six countries [49]. Inter-assessor agreement for the HCR-20 V3 is measured using intraclass correlation coefficients (ICCs). Studies have consistently shown high ICC values for the HCR-20 V3, with most ICC1 values ranging from 0.70 to 0.80 and most ICC2 values ranging from 0.80 to 0.90, mostly in the “excellent” or “almost perfect” range, and the lowest values were in the “good” range. The AUC values for HCR-20 V3 after 1-, 2-, and 3-year follow-up were 0.77, 0.75, and 0.67, respectively, with no significant difference in predictive validity compared to HCR-20 V2, and De Vogel et al. [50, 51] later followed up 78 women with AUC values of 0.71 and 0.67 at 3 and 12 years, respectively. These results demonstrate good reliability and validity of the HCR-20 V3.

As psychiatric rating tools are commonly used in risk assessment, the HCR-20 V3 has been included in correlation studies with other assessment tools. For instance, high correlations between the total scores of the HCR-20 V3 and the PCL-R have been observed [52]. Hogan and Olver [53] have examined correlations between the HCR-20 V3 and the Violence Risk Scale (VRS) and VRAG-R, revealing that the HCR-20 V3 total scale correlation scores are highly correlated with the VRS dynamic scale and total scale (0.77–0.83), slightly correlated with the VRS static scale and VRAG-R (0.56–0.69), with the H scale being highly correlated with both the VRS and VRAG-R (0.71–0.85) and the C scale being moderately correlated with the VRS dynamic and total scales (0.30–0.49) and not significantly correlated with the rest, and the R scale moderately correlated with both the VRS and VRAG-R, with moderate-to-large correlations (0.40–0.67). Furthermore, the HCR-20 V3 was also well correlated (0.52–0.66) with the Female Additional Manual, which can be used to complement the assessment of violence risk in women [38]. However, it remains to be demonstrated in subsequent studies whether the clinical use of HCR-20 V3 recommendations for risk management reduces violent behaviour in the assessed subjects.

The HCR-20-Chinese Version (HCR-CV) was translated and revised from the HCR-20 by Lv et al. [54] in 2015, with no significant changes to the overall subscales. An item assessing adverse family environment before the age of 16 years was added to the H scale. The research team also found that the removal of the PCL-R item from the H scale did not have a significant impact on the reliability of the HCR-CV. The research was conducted on 156 male patients with schizophrenia to verify the validity of the HCR-CV. The retest reliabilities for the total scale and subscales ranged from 0.523 to 0.953. The ICC values were 0.852, 0.965, 0.922, and 0.820 for total, H, C, and R scales, respectively. The AUC values at the 2-week follow-up were 0.721, 0.732, and 0.630 for H, C, and R scales, respectively, with a corresponding value of 0.852 for the total scale. At the 4-week follow-up, the AUC values were 0.883, 0.647, 0.624, and 0.820 for total, H, C, and R scales, respectively. However, statistical significance was only achieved for total and H scales across all statistical analyses. These findings suggest that the HCR-CV can provide comparatively accurate results for the short-term assessment of violence risk in hospitalized male schizophrenic patients.

Violence Risk Screening-10 (V-RISK-10)

V-RISK-10 is a tool that was specifically developed to assess the risk of violence among patients with general psychiatric disorders. Comprising four static and six dynamic violence risk factors, the tool is designed to screen patients who might require immediate further assessment or management. In 2011, Yao et al. [55] translated the V-RISK-10 into Chinese and validated its inter-assessor reliability and predictive validity. The study reported high ICC values for both reliability and predictive validity (0.89 for each). Additionally, the AUC values were reported as 0.62 and 0.63 for predictive validity, respectively.

Fourth-generation static/dynamic assessment stage

The fourth-generation violence risk assessment tools, which include both static and dynamic factors, were developed based on the third-generation tools to address specific types of violence, such as sexual violence, domestic violence, and youth violence. The fourth-generation tools allow for a more systematic assessment, management, and treatment of offenders, with a primary aim of identifying specific targets for intervention. These tools include the SORAG, the VRS, the SAVRY, and the Level of Service/Case Management Inventory (LS/CMI), among others.

SORAG

SORAG was developed from VRAG and was mainly used to assess the risk of recurring sexual or violent offences. Due to the higher-than-expected rate of violent recidivism among sexual offenders when assessed by the VRAG, a separate assessment tool is required to predict risk accurately [56]. The SORAG is a modification of the VRAG and consists of 14 items that can be scored in percentages. SORAG, similar to VRAG, can be scaled to produce a total score provided that no more than four items have missing or replacement scores [57].

Several studies have reported AUC values for the SORAG, which demonstrate its efficacy in predicting recidivism. The tool demonstrated AUC values of 0.73, 0.75, and 0.68 for violent recidivism, general recidivism, and sexual recidivism charges, respectively, with significant predictive validity for both the severity and timing of recidivism occurrence [58–60]. This scale has high predictive validity and is particularly important as a guide in determining whether sexual offenders who suffer from mental disorders can be released or whether they need mandatory medical treatment [61].

VRS/Violence Risk Scale-Sexual Offender version (VRS-SO)

The VRS is a fourth-generation violence risk assessment tool developed to identify appropriate treatment goals and assess treatment change, guided by the Risk-Need-Responsivity (RNR) and Psychology of Criminal Conduct theories. It was developed to meet the needs of the criminal justice system. The VRS consists of six static items and 20 dynamic items on a 4-point scale from 0 to 3, making it a comprehensive tool for assessing violence risk. The total score of the VRS is positively correlated with the level of violence risk, with the static items being scored independently of treatment interventions, while dynamic factors are accordingly changed. This combination allows for the assessment and prediction of violence risk, risk-reducing treatment, and treatment change. The RNR model, which is a risk demand response model, is widely regarded as the preferred model to guide offender assessment and treatment [62].

Several studies have assessed the inter-assessor reliability of the VRS, which yielded ICC values ranging from 0.96 to 0.98, 0.85 to 0.96, and 0.82 to 0.96 for static, dynamic, and total scales, respectively [63–66]. The predictive validity of the VRS has also been examined by scholars, who reported median AUC values of 0.75 [67–69].

The VRS-SO integrates sex offender risk assessment, treatment planning, and change assessment into a single instrument, which closely relates to the VRS. It comprises seven static and 17 dynamic factors. The reliability of the VRS-SO is similar to that of the VRS, with ICC values of 0.97, 0.74–0.92, and 0.86–0.90 for static, dynamic, and total scales, respectively. The median AUC values range from 0.67 to 0.72 [70–73].

The Violence Risk Scale-Chinese Version (VRS-C) was introduced to China with authorization by Zhang et al. [74] in 2011 and is mainly used to predict violence in psychiatric patients with criminal or violent behaviours and has been widely used in the forensic evaluation field. It has good reliability and validity and does not require a high degree of knowledge from the assessor. Thus, it can better meet the needs of forensic assessment. Similar to the VRS, the VRS-C has six static factors and 20 dynamic risk factors and employs a 4-point scale from 0 to 3. Chinese scholars evaluated the reliability of VRS-C scores by assessing 14 forensic psychiatric patients and conducting a homogeneity test with 125 patients with psychiatric disorders. The ICC values of static, dynamic, and total scales were 0.91, 0.70, and 0.80, respectively, which were all statistically significant [75]. Since the VRS was originally designed for in-prison populations and not limited to offenders with mental disorders, it is more appropriate for use in populations at moderate-to-high risk of violence. In China, it has been used with schizophrenic and non-psychotic populations suspected of homicide [76].

SAVRY

The SAVRY [77–80] is an SPJ risk assessment tool designed to evaluate the risk of violence in young individuals aged 12–18 years. The SAVRY consists of six protective factors and 24 risk factors and can also be used to assess the risk of general offending or recidivism. It is intended to provide advice on youth risk management or intervention plans. The SAVRY demonstrated high internal consistency and is a reliable tool in the assessment of youth violence risk. The assessor should be knowledgeable and cautious regarding their choice of the assessment tool. When assessing individuals aged 16–20 years, if the individual being assessed is an independent and autonomous individual, the assessor may prefer to use the HCR-20 V3. However, if the individual is less mature and dependent, the SAVRY may be a more suitable option. In the SAVRY assessment, a specific numerical score is not utilized. Instead, a protective factor presence/absence rating and a three-level rating of risk factors (high, medium, or low) are employed, which culminate in a summary judgement of the nature and extent of violence.

The reliability of the SAVRY has been assessed, with good inter-assessor agreement as indicated by ICCs ranging from 0.67 to 1.00 [81, 82]. Predictive validity studies of the SAVRY have reported AUC values mostly between 0.74 and 0.80 [83–86]. A meta-analysis of nine risk assessment tools has concluded that the SAVRY had the highest predictive validity based on statistical data [6]. Regarding the impact of gender differences on SAVRY assessments, while many studies reported that these differences are not evident, it has been suggested that the tool should be modified for girls because risk factors, such as sexual abuse, have different probabilities of occurring in males and females and cause different effects when they occur [87–89]. Furthermore, age and ethnicity might also impact the validity of the assessment, and further research is required to validate and refine its use.

Limitations and prospects

Currently, the importance of violence risk assessment has gained increased attention in China, leading to the development of various risk assessment tools for use in the clinical treatment or the judicial field. These tools are typically translated into Chinese and modified to better suit Chinese culture and customs, with the aim of providing a more objective and scientific basis for risk management or legal decisions. In contrast to other countries where violence risk assessment tools are commonly used to assess sentencing evidence in the legal field or develop individual risk management interventions in the clinical field, China primarily uses violence risk assessment within the context of compulsory medical treatment and determination of competence. Nevertheless, some assessment tools widely used abroad, such as the PCL-R, VRS-C, HCR-CV, and Violence Risk Screening-10, have been successfully implemented in China following localization and have demonstrated good reliability and internal consistency. However, assessors in China should exercise caution when selecting assessment tools, as their purpose and context of use can impact the validity of the results. For example, the MOAS may have limited value in predicting the risk of violence in clinical settings due to hospital constraints. Although the MOAS has a low value in predicting the occurrence of serious aggression within 1 month, it may still have some predictive value for a period of 2 years [90]. Therefore, assessors should consider these factors and select assessment tools with care.

The assessment of violence risk in patients with mental disorders is a complex task influenced by various factors that might affect the accuracy and reliability of assessments. These factors include differences in domestic and international laws and cultures, the context in which the person being assessed lives, the time frame for predicting the risk of violence, and varying levels of competence among assessors. For example, the existence of gun control policies in China, the lower crime rate in criminal cases compared to many countries, and the fact that many assaults are not necessarily outwardly displayed all contribute to the potential lack of reliability of the violence risk assessment. Furthermore, insufficient attention is paid to verbal assault in domestic assessments, which can easily lead to negligence. Scholars have conducted research in recent years to identify differences in biological markers, such as gut microbiota, biochemical indicators, and immune-related inflammatory factors, between groups with and without violence to better meet the need for objectivity and reliability in assessment results [91–95]. Additionally, they have built new models based on machine learning and data analysis to assess the possibility of violence in patients. These methods have allowed for more objective identification of violence risk in patients with mental disorders and timely interventions.

Currently, violence assessment methods in China are mainly used in forensic institutions to guide the assessment of violence risk of offenders with mental disorders, and to a lesser extent in specialized psychiatric institutions (including compulsory medical treatment facilities) to predict and prevent aggression in hospitalized patients [96–98]. Risk assessment of violence in custodial institutions, such as prisons, detention centres, and drug rehabilitation centres, requires focused attention due to the unique nature of these institutions, where offenders with extreme and repeated violent behaviour are housed [46, 99]. However, the manifestation of violence in custodial institutions is much more difficult to observe than in the community or hospitals due to the contextual factors involved. Schug et al. [100] suggested the use of a multilevel assessment combining three assessment techniques to construct possible personality bases associated with violent behaviour to accurately assess the risk of violence in offenders with extreme and repeated violent behaviour. This method was refined through case examples to assist in future risk assessment and clinical treatment. To enhance the accuracy of risk assessment in custodial institutions, it might be necessary to rely on certain emotional arousals, such as arousal in high-pressure situations, as the risk of violence is assessed through the analysis of micro-expressions that are not under autonomous control. Micro-expression recognition methods have been gradually applied to determine mood changes in several fields, including criminal investigation and applied psychology, and they combine physiological manifestations such as heart rate, voice, and eye movements. These methods have helped improve the accuracy of assessment and have achieved some success in lie detection. However, their reliability is still limited by insufficient public databases and few large-scale data labels, which calls for further research to enhance their reliability [101].

In conclusion, the accuracy of violence risk assessment tool results might be influenced by various internal and external conditions. The ICC values of static factors in the assessment tool items are greater than those of dynamic factors. However, dynamic factors are often more important in managing violence risk and intervention treatment. Static factors are less susceptible to assessor competence, treatment, or other factors. Therefore, optimization efforts should focus on dynamic factors. In addition to optimizing the items of the tool, future research should also concentrate on developing new tools based on data analysis and objective assessment criteria, such as biological markers. These tools should serve as adjuncts to scale assessment and be supported by databases to analyse components that assessors might easily overlook. This can reduce the influence of subjective assessor factors and objective contextual factors on assessment results to a certain extent. Consequently, violence risk assessment can be performed more scientifically and objectively in the future.

Conclusion

Violence risk assessment plays a significant role in both psychiatric clinical treatment and forensic assessment. Optimizing violence risk assessment methods is essential to enhance the accuracy of assessments, reduce inter-assessor variability, and provide more objective and scientific results. However, external factors can affect the accuracy of the assessment methods, making it crucial to improve and develop new methods. In China, the main purpose of violence risk assessment is to determine responsibility in legal cases and mandate clinical treatment, which differs from overseas practices. Therefore, it is necessary to tailor domestic violence risk assessment methods to suit the specific needs and conditions of the country. Facial micro-expression, voice, handwriting, and other external behaviours and biological markers associated with violence risk assessment should be considered when developing new methods. The development of new methods and the consideration of various external factors can lead to more reliable and scientific assessments. Moreover, the use of artificial intelligence for data analysis can further enhance the accuracy of the assessment methods, reducing discrepancies caused by subjective judgements.

Authors’ contributions

Xindi Ling and Haozhe Li participated in the design of the study and drafted the manuscript. Xindi Ling, Haozhe Li, Wen Li, Shujian Wang, and Qinting Zhang collected and analysed data. Haozhe Li and Weixiong Cai conceived the study protocol and participated in the coordination of the study. All authors contributed to the final text and approved it.

Compliance with ethical standards

This article does not contain any studies with human participants or animals performed by any of the authors.

Disclosure statement

None declared.

Funding

This study was supported by the National Natural Science Foundation of China [grant number 81801881], Science and Technology Committee of Shanghai Municipality [grant numbers 20DZ1200300, 21DZ2270800, and 19DZ2292700], and Ministry of Finance, PRC [grant numbers GY2025G-5, GY2023Z-4, and GY2020G-3].