Prediction of Risk of Harm of Inpatient Aggressive Behaviours in St Joseph’s Healthcare Hamilton

Abstract

Inpatient violence risk prediction is a priority for safety and quality control purposes. The aims of this retrospective study is: 1) analysing the performance of the Risk of Harm to Others Clinical Assessment Protocol (RHO CAP) in predicting the risk of harm within St Joseph’s Healthcare Hamilton; 2) identifying the most important risk factors associated with harmful incidents in inpatient mental health; 3) developing an alternative algorithm to predict the risk of harm in inpatient settings; 4) analysing the performance of the RHO CAP among patients who did not commit any aggression. Data from January 2016 to December 2017 have been anonymized and collected, for a total of 870 episodes of inpatient aggressions perpetrated by 337 patients. Two main sources of information have been used: the Resident Assessments Instrument for Mental Health (RAI-MH), and an internal report of the aggression incidents. We develop a Bayesian probabilistic classifier, a logistic regression, to investigate the risk factors for harmful incidents and propose a predictive model for risk of harm. The RHO CAP has demonstrated a better performance in discriminating which patients were more at risk to commit some type of aggression than at identifying the risk of harm among those who will commit aggression. The factors most significantly associated with harmful incidents were age, history of violence to others, police intervention for violent behaviour, and a diagnosis of psychosis. The proposed predictive model showed an overall accuracy of 75%. It focuses on four diagnoses (the most frequent diagnoses associated to aggression in our sample), age, history of violence to others and police intervention for violent behaviour, and inappropriate behaviour within the social context.