Data Science
Artificial intelligence for dementia
The challenge
The use of artificial intelligence and machine learning hold significant promise in the field of dementia research, but their utility and efficacy to date has not been critically appraised and placed in context. How can the field best progress towards the use of artificial intelligence and associated technologies, with regards research and clinical use?
The research
As part of a commission from the Alzheimer’s Research UK (ARUK) deep phenotyping network (DEMON), we appraised all existing studies into the use of digital health (e.g. online assessment, objective measurement of physical activity, apps), and machine learning in the prediction and assessment of dementia. The project was highly multi-disciplinary, including psychologists, computing scientists, epidemiologists, geriatric clinicians and experts in digital assessment (e.g. accelerometers and gait).
The results
As larger cohorts provide more detailed and rich data (e.g. proteomics), the importance of machine learning in decomposing large numbers of variables, into fewer, critical factors, will become key. Naturalistic, objective data sources hold significant promise in understanding behaviours that predict, and are changed in, dementia. This is because self-report of some things, like aspects of sleep and physical activity, are not necessarily consistently reliable. As it stands, there are complex ethical and practical considerations before predictive modelling can reliably benefit ‘real world’ clinical management of dementia at the individual, patient level.
The impact
We provide several recommendations for the field going forward, including more streamlined data access for researchers, use of naturalistic data assessment (e.g. accelerometers for sleep health), and a move beyond simple ‘prediction’ of dementia towards more fine-grained outcomes (e.g. progression from mild to severe impairment, or predicting age of disease onset).
Lead
Read the paper
The study, Artifical intelligence for dementia - Applied models and digital health, is published in Alzheimer's & Dementia.
