Little evidence for the usefulness of violence risk assessment

Troquete and colleagues report a cluster randomised trial of the effect of violence risk assessment on future offending.¹ They found that people in the risk assessment group were non-significantly more likely to re-offend than those in the control group. We welcome this analysis of the practical value of risk assessment. There are now literally thousands of published violence risk assessment studies, most of which claim validity for their risk assessment method on the basis of statistical discrimination between violent and non-violent groups using measures such as the area under the curve (AUC) or other indicators of effect size.² Recent criticism of the AUC as an outcome measure has emerged because it does not reflect the accuracy of predictions in the real world, and even high AUC values are associated with a low positive predictive value (PPV) for rare events. However, the PPV of a risk assessment is only a proxy for the usefulness of a risk assessment. A risk assessment alone is not valuable unless it leads reasonable interventions that can reduce future harm. Therefore, the utility of a risk assessment must ultimately be judged by its ability to contribute to harm reduction. In contrast to the large number of papers about the statistical aspects of risk assessment, there may be as few as four published controlled studies of the ability of risk assessment to reduce harm.²

The British Journal of Psychiatry has published two earlier studies of the utility of risk assessment. Abderhalden et al reported a cluster randomised trial of risk assessment among in-patients that found that intervention wards had a reduction in violence. However, interpretation of this study is difficult because the intervention wards had high rates of violence pre-trial and post-trial rates of violence in the experimental and control wards did not differ.³ Also in the Journal, van de Sande and colleagues reported a cluster randomised trial that found that risk assessment was associated with a reduction in violence but not seclusion among in-patients.⁴ In the nursing literature, Kling et al reported a study in in-patient settings that found that risk assessment was not helpful in reducing violence.⁵

Risk assessment has become the dominant paradigm in mental health practice, policy and legislation in most high-income countries. It should therefore trouble colleagues who support ‘evidence-based practice’ to know that there is so little evidence for the effectiveness of risk assessment.