| Summary: | Introduction. Emotional dysregulation often occurs in people with an acquired brain injury (ABI), and can lead to challenging behaviour including aggression. Emotion regulation is associated with the autonomic nervous system and physiological recordings (e.g. heart rate variability) can be used as index measures of self-regulation. The development of wearable devices allows for real-time continuous physiological recording. Advances in machine learning techniques also open avenues for analysing these data, which involves identifying patterns to predict behaviours. This study aimed to explore whether physiological and sleep recordings could be used to accurately predict challenging behaviour in individuals with an ABI. Methods. Participants were recruited from a brain injury inpatient unit. Participants wore a Smartwatch, which collected physiological and sleep data. Staff recorded episodes of verbal and physical aggression. Ethical approval was obtained for the study. Data mining techniques were used to develop models for predicting aggressive behaviour. Results. Five participants were included in the study. Technical and practical problems led to unanticipated data collection difficulties, including participants failing to wear or charge the devices, smartwatches breaking, and poor WiFi connection. Machine learning was used to create predictive models both for individual participants, and a combined model. Models successfully predicted 9-100% or episodes of aggressive behaviour however there were a large number of false alarms impacting the clinical applicability of the models. Discussion. This study failed to create a model that predicted episodes of aggression without a sufficiently low number of false alarms that would suggest it was clinically useful in an inpatient setting. Several possible reasons are discussed. Practical recommendations for future research include a careful choice of smartwatch,a reliable internet connection, consistent aggression reporting, and the inclusion of a self-reported measure of emotional dysregulation. Conclusions. The development of a clinical tool that can accurately predict and warn individuals, or staff, of imminent emotional dysregulation opens avenues for its prevention. This study provided useful insight into the initial attempts to explore this using a machine learning approach in the ABI population.
|
|---|
