OBJECTIVE: The authors compared the predictive accuracy of two risk assessment methods that are feasible to use in routine clinical settings: brief risk assessment tools and patients' self-perceptions of risk.

METHODS: In 2002-2003, clinical interviewers met with 86 high-risk inpatients with co-occurring mental and substance use disorders (excluding schizophrenia) to carefully elicit the patients' global rating of their risk of behaving violently and to complete two brief risk assessment tools-the Clinically Feasible Iterative Classification Tree (ICT-CF) and the Modified Screening Tool (MST). Two months after discharge, patients were reinterviewed in the community to assess their involvement in violence.

RESULTS: Patients' self-perceptions of risk performed quite well in predicting serious violence (area under the curve [AUC]=.74, sensitivity=50%), particularly compared with the ICT-CF (AUC=.59, sensitivity=40%) and the MST (AUC=.66, sensitivity=30%). Self-perceived risk also added significant incremental utility to these tools in predicting violence.

CONCLUSIONS: Patients' self-perceptions hold promise as a method for improving risk assessment in routine clinical settings. Assuming it replicates and generalizes beyond the research context, this finding encourages a shift away from unaided clinical judgment toward a feasible method of risk assessment built on patient collaboration.