A prospective study of predictors of violence in adult male mental health admissions.

Studies of violence prediction have typically involved long-range follow-up of institutionalized persons and have shown poor predictive accuracy. In the present study a sample of adult males at risk for violent behavior admitted as inpatients at a community mental health center was followed for six months after their release. Data collected at admission (which included demographics, family back-ground, criminal justice and mental health systems contacts, past violent behavior, life stress, family, friendship, and work environment measures, psychiatric diagnosis, and alcohol and drug use) were used to predict subsequent violent arrests or readmissions using stepwise discriminant analysis. Results yielded 85% of the total sample correctly classified. Of those predicted to be nonviolent 94% were actually nonviolent and of those predicted to be violen 59% were actually violent. Seventy-six percent of the violent subjects were identified. Implications of the findings for clinical predictions of violence are discussed.