The Predictive Properties of Violence Risk Instruments May Increase by Adding Items Assessing Sleep

Background: The psychometric instruments developed for short-term prediction of violence in psychiatric inpatients do not include variables assessing sleep. Disturbances in sleep may precede aggression in this setting. We investigated whether adding information on sleep improved the predictive properties of the Brøset Violence Checklist (BVC). Methods: The study population consists of all patients admitted to a psychiatric intensive care unit (PICU) over a 6-month period who were hospitalized for at least one night (n = 50). Sleep observed by staff (521 nights), behavior assessed with the BVC (433 days), and aggressive incidents recorded by the Staff Observation Scale-Revised (n = 14) were included in the analysis. Results: The ability of the BVC to predict aggressive incidents improved from AUC(ROC) 0.757 to AUC(ROC) 0.873 when a combined sleep variable including both sleep duration and night-to-night variations of sleep duration was added to the BVC recordings. The combined sleep variable did not significantly predict aggressive incidents (AUC(ROC) 0.653, p = 0.051). Conclusions: A sleep disturbance variable improves the predictive properties of the BVC in PICUs. Further studies of sleep duration, night-to-night variations in duration of sleep, and aggression are needed.