The HCR-20 is the most widely used structured violence risk assessment tool in medium secure units in England and has good predictive validity for inpatient violence. However little is known about whether its predictive validity varies between different clinical and demographic groups.

1. 1) collect HCR-20 risk assessment and three month follow-up data for all eligible inpatients

2. 2) examine the ability of the HCR-20 to predict various indices of inpatient aggression using receiver operating characteristics (ROC), and

3. 3) compare AUC values derived from different groups (i.e. men vs. women)

determine if the HCR-20 performs equally across different clinical and demographic groups.

This study was conducted amongst inpatients at St Andrew's (N=505). Demographic (age, ethnicity, gender), clinical (diagnosis) and routinely collected risk assessment data (HCR-20) were collated. Incidents of aggression were coded using the OAS for the three months following assessment.

ROC analysis revealed that the HCR-20 total, Clinical scale and Risk Management scale were predictive of all types of aggression; the Historical scale was a significant predictor of verbal aggression and physical aggression towards objects but not physical aggression towards people. The predictive efficacy of the HCR-20 differed according to gender and diagnosis, with superior performance obtained with women and those with a diagnosis of personality disorder.

This study reveals differences in the HCR-20's efficacy between groups. Further development and optimisation of the HCR-20 may be required. Alternatively, interventions to prevent aggression may be more effective among some groups. Future research should address these issues.