Risk assessment and clinical decision-making

To the Editor

In a recent commentary, Large et al. (2014) argue that violence risk assessments do not lead to a better clinical praxis, but rather the opposite. The main arguments are that a categorical application of risk assessments might lead to: (i) a large group of false positive predictions which could allocate resources from treatment to unnecessary restrictions or reports; (ii) non-treatment among low-risk patients as a result of low-risk categorization, which could lead to violent episodes; and (iii) accuracy being insufficient. Their basis was systematic reviews of the literature, consisting mainly of risk prediction studies.

One problem with risk prediction studies in psychiatric settings is that for obvious ethical reasons staff and other involved personnel have to prevent violent behavior in both inpatients and outpatients, and not just observe if it happens. Good risk assessment and good treatment/management could lead to the prevention of violence, and a good risk assessment could then become a wrong risk prediction. Some of the false positive predictions might be ‘iatrogenic’ due to good risk management. The better the risk management, the more ‘iatrogenic’ false positive predictions there are, and the lower the accuracy for risk assessments and predictions.

Studies of the accuracy of risk predictions may not be the best way to evaluate risk assessment instruments. Pre-post studies explore whether implementation of a risk instrument might lead to diminished violence or restraints. In a semi-randomized cluster trial of 14 acute departments (the departments were randomized) in Switzerland, Abderhalden et al. (2008) found a 41% reduction of violent episodes and a 27% reduction in the use of restraints after implementing a short-time risk assessment instrument (Brøset Violence Checklist – BVC) in seven of the wards, compared with the seven wards giving ‘treatment as usual’. In a randomized cluster trial, van de Sande et al. (2011) found a similar reduction in inpatient violence and time spent in seclusion after implementation of BVC in two wards compared with two wards giving ‘treatment as usual’. The most important aspect, and more important than predictive accuracy, is whether risk assessments lead to decreased violence. Large et al. (2014) seem to have overlooked the pre-post studies, which also should be emphasized when the use or usefulness of risk assessments are discussed.