Personality disorder and violent criminality : a follow-up study with special reference to psychopathy and risk assessment

Abstract: Background: Structured models for the assessment of psychopathy and risk of violent recidivism have gained currency in forensic and correctional practice in many countries. The overall aim of this study was to test the predictive validity of some of these models among violent offenders with personality disorders in Sweden. Method: The study was a retrospective follow-up of all offender convicted of a violent crime and subjected to a forensic psychiatric evaluation during 1988 - 90 diagnosed with a personality disorder with or without concomitant substance disorder (N=401). Data on background, clinical and crime characteristics as well as recidivism during follow-up were obtained from archives and registers. Predictive validity was estimated with receiver operating characteristic (ROC) analysis. Results: In accordance with the current Swedish legislation, the vast ma)'ority of offenders with PDs are sentenced to prison instead of treatment (Paper I. Methodological pre-studies on the archive and register material yielded a set of historical, crime-related, and clinical variables that could be reliably extracted for research purposes (Papers II-III), including psychopathy according to Hare's Psychopathy Checklist - Revised (PCL-R). Psychopathy was found to be clearly associated with an increased risk of violent recidivism, and was demonstrated to have a moderate, statistically significant predictive validity (Paper IV). This was also the case with two actuarial risk assessment instruments; the Violence Risk Appraisal Guide (VRAG) and the historical part of the so-called HCR-20. In terms of ROC-curves, these instruments discriminated between recidivists and non-recidivists with areas under curves of .68 and .71 (Paper V). The largest ROC-area (.81) was found in a study in which the risk for recidivism was modeled with an artificial neural network (Paper IV). Conclusions: The results of this study suggest that these models exhibit a clear, albeit moderate, predictive validity in the Swedish context as well. Future research may benefit from prospective study design, inclusion of other offender groups, and broader outcome measures.