Objective: This study sought to assess the validity of different combinations of readily available clinical information in predicting assaults by patients with psychosis, predominantly in the community. The combinations of information were: a) age and sex, b) age, sex and history of criminality/violence c) age, sex, history of violence and drug use and d) age, sex, history of violence, drug use and personality disorder.
Method: 708 subjects were followed for 2 years. Assaults were measured using multiple sources of information. Prediction validity was measured using the area under the receiver operating curves (AUC) and the number needed to detain (NND). A simple prediction tool was developed.
Results: The AUC values using the four combinations of information were a) 0.65, b) 0.70, c) 0.71, and d) 0.73. Prediction based on combination b), c), and d) implied a NND of 3. A rule based on c), the most accessible information, is suggested as a simple screening tool.
Conclusions: Readily available clinical information allowed the prediction of assault over 2 years, in a sample of general psychiatric patients with psychosis, with a level of predictive accuracy comparable to that described using more detailed risk assessment tools. The information used in the predictive model was: age, sex, having committed an assault in the last 2 years (self-report) and having used any drug in the last year (self-report).