BACKGROUND: Male-to-female spousal violence is a significant public health problem because of the associated physical, psychological, and financial costs to victims, their families, communities, and society in general. Worldwide 10% to 69% of women reported being physically assaulted by an intimate male partner at some point in their lives. The 2004 General Social Survey indicated that approximately 7% of Canadian women 15 years of age and over in a current, previous, or common-law union experienced spousal violence in the previous five years. To prevent future serious or lethal assaults and to help allocate already scarce resources, risk assessment instruments have been developed in an attempt to identify offenders who are at high risk of reoffending. However, little is known about the predictive validity of these instruments.
OBJECTIVE: To assess the research evidence on the inter-rater reliability and predictive validity of various risk assessment instruments in predicting male-to-female spousal violence recidivism and lethality in those males who had contact with the police system.
METHODOLOGY: All original published systematic reviews or primary studies were identified by systematically searching PubMed, CRD databases (NHS EED, HTA, DARE), EMBASE, Family & Society Studies Worldwide, Sociological Abstracts, Social Services Abstracts, Social Sciences Abstracts, Academic Search Premier, Web of Science, PsycINFO, and ABI/Inform from January 1995 to May 2007. Relevant library collections and websites of health technology assessment-related agency resources were also searched.
RESULTS: No systematic review on this topic was located through a comprehensive literature search. Eight primary studies were found that evaluated the predictive validity of several currently used instruments, including the Ontario Domestic Assault Risk Assessment (ODARA), the Spousal Assault Risk Assessment (SARA), the Danger Assessment (DA), the Domestic Violence Screening Instrument (DVSI), the Violence Risk Appraisal Guide (VRAG), and the Level of Service Inventory-Revised (LSI-R). The ODARA was evaluated in only one Canadian study, whereas the SARA was evaluated in five studies. The characteristics of the population under study—male offenders—varied considerably across the included studies. Some male offenders were arrested, on probation, or in a maximum-security psychiatric facility, whereas others were referred to attend batterer treatment programs that were provided in Canada or the United States. Inter-rater reliability was tested for the SARA, the ODARA, and the VRAG, but not for the DA and other instruments. Limited research indicated good inter-rater reliability for the SARA, the ODARA, and the VRAG. In terms of predictive validity, five studies compared the predictive validity of one instrument with other instruments, whereas three studies reported the predictive validity of only one instrument. None of the studies reported any lethal assault during the follow up period. The reported area under the receiver operating characteristic curve (ROC AUC, a measure of predictive validity) was less than 0.80 (ranged from 0.59 to 0.77) for all instruments under evaluation, suggesting only marginal to moderate improvement over chance in predicting non-lethal recidivism. Only one study that was conducted in Canada compared predictive validity of the ODARA, the SARA, and the DA. On the basis of the ROC analysis, the predictive validity of the ODARA appeared to be lower in the cross-validation sample than that in the construction sample. In the construction sample, all three instruments yielded statistically significant predictive validity, but the ODARA predicted recidivism statistically significantly better than did the SARA or the DA. In the cross-validation sample, the ODARA significantly predicted recidivism, whereas the DA and the SARA did not. These findings need to be interpreted cautiously. First, instructions for the SARA were not completely followed; that is, no interview was performed, and no clinical judgment was made. Second, no lethal assault occurred during the follow up in this study; therefore, it could not be concluded that the ODARA was superior to the DA in predicting lethal assault.