AI and Machine Learning for Detection and Management of Delirium in Care Home Residents

Background: Presently, diagnosing delirium in older people is a challenge. Diagnostic support tools such as the Confusion Assessment Method and 4AT provide structure but require specialist training, resources, and implementation support, while some subjectivity persists in diagnosis. This is particularly the case in people who live with dementia who often experience rapid fluctuation in cognitive abilities and behaviours. This leads to variation in diagnosis between settings and care providers, with consequent harmful impact on those experiencing delirium. These challenges become greater in care homes where dementia is prevalent, daily fluctuation is the norm, and the majority of staff are not trained healthcare professionals. Summary: Here, we outline the potential for AI-based human activity recognition (HAR) approaches to identify and flag deviations from normal behaviour that may be precursors of a delirium state, enabling earlier detection and management, and better outcomes. We outline how statistical process control approaches could form the basis of diagnostic algorithms and the steps required to test the feasibility of this approach in the care home setting. Key Messages: Delirium detection and diagnosis, difficult in any setting, are more difficult in care homes because of resident, staff, and organisational factors. Artificial intelligence, machine learning, and HAR have potential to make diagnosis more reliable because of their ability to recognise changes from normal patterns of behaviour at an individual level.