Short-term violence risk assessment in acute psychiatry – A naturalistic prospective inpatient study

Background: This doctoral thesis is based on a PhD project that was part of a larger prospective naturalistic research project: “Violence risk assessment in an acute psychiatric ward”. Violence within the health sector is a global and increasing concern. Aggression and violence occurs frequently in mental healthcare settings and is perpetrated by both males and females. Current violence risk assessment methods seem to have reached an upper limit of accuracy. A broader approach to risk assessments is suggested, hence different perspectives may provide a deeper and improved understanding of patients’ violence risk. Risk assessments in acute psychiatric settings must also be performed quickly, and predictions must be accurate over the short-term. Aim: The overall aim of this doctoral thesis was to investigate an extended model for shortterm violence risk assessments in acute psychiatric units. The thesis investigated patients’ self-assessments of risk (SRS) as a risk marker for inpatient violence (Study 1), the association between short-term risk assessments with the Brøset Violence Checklist (BVC) and imminent violence throughout hospitalisation (Study 2), and whether an extended model combining BVC, SRS and single items from Violence Risk Screening 10 (V-RISK-10) provides improved short-term predictive accuracy compared to the BVC alone (Study 3). Methods: The target population was all patients involuntarily and voluntarily admitted to an acute psychiatric ward in Oslo over one year between March 21, 2012, and March 20, 2013 (N = 558). Thirty patients withdrew from participation, resulting in an initial study sample of 528 patients from 717 admissions. In Study 1, logistic regression was conducted to investigate the predictive accuracy of SRS. Interaction analysis was used to investigate gender differences. In Study 2, logistic regression and generalised linear mixed model (GLMM) analyses were conducted to investigate the predictive accuracy of the BVC. Interaction analyses were conducted to investigate possible gender differences. In Study 3, stepwise multivariate GLMM analyses were conducted: (i) BVC, (ii) BVC + SRS. (iii) BVC + SRS + single items from V-RISK-10. Findings: Patients who reported their risk of violence to be either moderate, high, don’t know or refused to answer (positive predictor of SRS) were more than four and a half times more likely to be violent, compared with those who reported no or low risk. Interaction analysis showed significant gender differences with SRS as a stronger risk predictor for women (Study 1). Findings confirm that the BVC is a suitable short-term risk assessment instrument throughout hospitalisation in acute psychiatric units, also when differentiating threats and physical violence, differentiating males and females and adjusted for diagnostic subpopulations and circadian variability. Interaction analyses displayed no significant differences in the association between males and females (Study 2). When adjusting for repeated measurements, an extended model for short-term risk assessment, consisting of BVC, SRS and Item 2 Previous and/or current threats in V-RISK-10 explained more of the imminent violence, compared to the BVC alone (Study 3). Conclusions: This is the first attempt to investigate an extended model for short-term risk assessments by including inpatients’ own risk assessments as part of the model. Inpatients’ self-assessments may contribute as a risk marker for inpatient violence. The findings confirm the predictive validity of BVC when adjusted for repeated measurements. Findings also indicate that an extended model for short-term risk assessments explains more variance of imminent violence than short-term risk assessments with the BVC alone. Nonetheless, the results must be interpreted with caution, and a number of limitations should be borne in mind