Use of risk assessment instruments to predict violence in forensic inpatient settings: a systematic review and meta-analysis

BACKGROUND & AIMS: Violent behaviour by psychiatric inpatients can have many negative consequences for the physical and mental health of both psychiatric staff and other inpatients. The problem is of particular concern on forensic psychiatric wards, where the prevalence of violence is higher than on other inpatient wards. In order to assess, manage and potentially prevent this type of violence, instruments have been developed with the aim of identifying individuals who are at increased risk of violence within a certain timeframe. This systematic review and meta-analysis aims to investigate the accuracy of these risk assessment tools for the prediction of violence on forensic inpatient wards. METHODS: The nine most commonly used violence risk assessment instruments that have been used in inpatient settings were included. A systematic search of five databases (CINAHL, Embase, Global Health, PsycINFO and PubMed) was conducted to retrieve all studies examining the predictive accuracy of these tools in forensic inpatient settings. A range of accuracy estimates and descriptive study- and sample-related variables were extracted. A quality assessment was performed for each eligible study using the QUADAS-2 (a tool designed to assess methodological quality for systematic reviews of studies investigating diagnostic or prognostic accuracy). Summary performance measures and HSROC curves were produced and meta-regression analyses investigated study and sample effects on accuracy. RESULTS: Fifty-two eligible publications were identified, of which 43 provided information on tool accuracy in the form of AUC statistics. These provided data on 78 samples, with 7,705 participants. Due to lack of sufficient data reporting, 35 samples (3,306 participants from 19 publications) could be included in the full meta-analysis of all performance measures. Risk assessment instruments were separated into those designed for imminent violence prediction and those designed for longer-term prediction (such as the HCR-20 and PCL-R). Imminent tools performed well for the screening out of low risk individuals, with a summary specificity of 0.99 (95% confidence interval [CI]: 0.80-1.00) and negative predictive value (NPV) of 0.99 (interquartile range [IQR]: 0.85-1.00). Identification of higher-risk patients was poorer, with a summary sensitivity of 0.59 (95% CI: 0.29-0.83) and a median positive predictive value (PPV) of 0.36 (IQR: 0.10-0.93). For longer-term tools, the summary accuracy estimates were as follows: sensitivity = 0.75 (95% CI: 0.65-0.83); specificity = 0.56 (95% CI: 0.46-0.66); PPV = 0.56 (IQR: 0.30-0.75); and NPV = 0.75 (IQR: 0.58-0.95). As an overall measure of accuracy, the median area under the curve (AUC) value for the wider group of 78 samples indicated better performance for imminent tools â AUC = 0.83 (IQR: 0.71-0.85) â compared with longer-term tools with an AUC of 0.68 (IQR: 0.62-0.75). Meta-regression analyses indicated that no study- or sample-related variables were associated with between-study differences in AUCs. DISCUSSION & CONCLUSIONS: This is a systematic review and meta-analysis of the predictive accuracy of violence risk assessment instruments for forensic inpatient violence. Imminent risk assessment instruments, including the Brøset Violence Checklist (BVC) and the Dynamic Appraisal of Situational Aggression (DASA), were found to be more accurate for the prediction of inpatient violence in forensic wards than those designed for longer-term prediction, such as the HCR-20 and VRAG.