Drivers of Seclusion and Physical Restraint in an Acute Mental Health Unit: A Feature Analysis

Understanding the drivers of seclusion and physical restraint supports the work towards minimising their use in acute mental health units. However, evidence on their most important drivers remains limited and is focused mainly on individual-level features. Employing 249 days of 917 contemporaneous records of nurse de-escalation events in one adult inpatient unit in regional Australia, from January 2019 to March 2020, twenty-three features other than individual demographic, dispositional, and diagnostic factors were extracted. Bivariate statistics and supervised machine learning algorithms for feature selection (i.e. Boruta algorithm) and predictive modelling (i.e. random forest) were applied. Emerging top drivers include incidents in high observation beds, the assessed level of situational aggression before de-escalation, incidents directed towards nurses, verbal de-escalation, and distraction and redirection. These findings elevate the predictive value of contextual and interventional, rather than individual-level, features in understanding the likelihood of restrictive practices.