Background: Persons living with dementia (PLwD) exhibit behavioral and psychological symptoms, with agitation and aggression being the most common. Aggression is associated with many risks, including psychological trauma or physical harm to PLwD or to caregivers. Traditionally, agitation is measured using retrospective behavioral scales such as the Neuropsychiatric Inventory (NPI) and Cohen-Mansfield Agitation Inventory (CMAI). These measurements require observation and data collection by health care providers whose time and intellectual labor is resource intensive. These scales are also prone to errors in recall and user bias, which impacts on the reliability and validity of results. To help address these issues, previous researchers have used actigraphy to detect incidences of agitation and aggression in PLwD. However, actigraphy-based solutions are limited to body movement based parameters.

Purpose: To address the limitations of previous practices, our research team developed a novel multi-modal sensing framework that has been installed and tested at Toronto Rehabilitation Institute, Canada.

Methods: DAAD, a framework to Detect and predict incidences of Agitation and Aggression in persons living with Dementia, uses video cameras, wearable device (for both movement and physiological data), motion and door sensors, and pressure mats to collect various types of data. To ground this sensor data, we have developed a documentation protocol to record patient’s responsive behaviors from the healthcare staff clinical documentations (located in patient’s charts) that consist of the time, nature of the patient’s responsive behaviour and its location on the unit. This documentation strategy helps find the corresponding camera and sensor data, and locate the exact timelines of agitation episodes. Therefore, the agitation and aggression episodes documented in the patient’s chart by the healthcare staff are used as the gold standard for annotation of video and sensor data. This step will help in improving the quality of the labelled data and is essential in building accurate machine learning classifiers to automatically identify and predict incidences of agitation and aggression.

Preliminary Results and Next Steps: The clinical trial is underway and to date, we have recruited two PLwD. The DAAD sensor framework has been able to seamlessly collect data successfully without technical glitches. We plan to recruit at least eight more PLwD over the upcoming months and begin analyzing the collected data.

Implications: The study represents the first steps towards developing a predictive system for agitated behaviours in dementia, which has the potential to be deployed in the home of PLwD, in institutions, and in hospitals. It has the potential to improve care and safety for caregivers in the home and in other care settings. This study represents a necessary phase toward the development of evidence-informed policy for improving predictors of behavioural episodes through a streamlined process of secure behavioural documentation for enhanced clinical decision-making.

We would like to acknowledge the Alzheimer’s Society Research Program Grant, Faculty of Community Services, Ryerson University, and University of Toronto for their support to conduct this research. Also, we would like to acknowledge Toronto Rehabilitation Institute participants and healthcare providers who were involved in the study.