Application Title Development of a disease prediction model aimed to assess the 10-year risk of dementia based on sociodemographic, health, lifestyle, cognitive data, biomarkers and genomics.
Machine learning (ML) is a method of computer programming that learns from data to make predictions about a specified outcome. In this study, we aim to build ML models (from existing data patterns) that can differentiate between healthy ageing individuals and those at risk of dementia. The model will predict the likelihood of developing dementia based on data from sociodemographic information (information about the structure and social details of populations), health and lifestyle questions, brain function (cognitive) assessment, genetics and blood biomarkers (characteristics in the blood which are a sign of disease or illness).
Developing such ML models will initially involve identifying patterns in sociodemographic, health and lifestyle data from UK Biobank, such as education, smoking, BMI (body mass index), diabetes, blood pressure and depression to build a prediction (evaluation) model that can assess the likelihood that someone will develop dementia within 10 years. At the second stage, we will investigate if and how cognitive data (information about memory, processing speed and thinking abilities) can add value to the accuracy of the ML model developed, and lastly, we will do the same with blood biomarkers and genetic data. We intend to use the ML model in our product (mobile device application, Sharp/ Five Lives). This will then be used for validation research for our newly designed cognitive tests that assess memory, processing speed etc. We expect that the project duration will last approximately 12 months.
The number of people suffering from dementia is increasing worldwide with no effective treatments available.Therefore, methods to prevent and delay the onset of dementia, as well as an increased understanding of the factors that can increase the risk of developing dementia are crucial to improving the management of dementia cases. Being able to identify individuals who are at an increased risk of dementia will allow for more efficient targeting of available measures that prevent the development of dementia than what is currently possible.
The potential use of digital biomarkers (behavioural or physiological measures collected from digital devices) within prediction tools to detect the early signs of existing and future cognitive decline is also a growing topic of research in the field. Previous studies showed that computerised memory games could detect early signs of cognitive decline. The UK Biobank database also holds cognitive assessment data and we will analyse this data with the aim to improve our prediction ML model.
The aim of this research study is to develop a machine learning (ML) algorithm to predict the risk of dementia in the next 10 years. We aim to develop this algorithm using sociodemographic, health, lifestyle, cognitive, biomarkers and genomics data held in the UK Biobank resource. Predictive models to date do not integrate cognitive data. We believe that this additional data will help to have a higher predictive accuracy in our model than existing models, such as the CAIDE (Kivipelto et al., 2006), ANU-ADRI (Anstey et al., 2013) and DRS (Walters et al., 2016).
Our overall aim is to develop a mobile application, 'Sharp', that assesses the risk of dementia among healthy people who are worried about dementia and their brain health, aged 50+. The 'Sharp' app will predict the user's risk using a ML algorithm that combines various risk factors and cognitive data to generate a risk score. Our objective is to build this risk prediction model supported by multimodal data from the UK Biobank.
Sharp is a Class I medical device (CE mark). The company, SharpTx, is a DPUK partner with world-class scientific and medical advisors from leading universities which follow ISMS processes.
We would like to explore the possibility of developing algorithms which can predict dementia up to 20 years before dementia onset. Additionally, based on initial analyses of our users' data we have identified the need to quantify the current likelihood of them having MCI or dementia in the interest of being able to safely sign-post those that may benefit from formal clinical input. Additionally, as we are aiming to provide a digital coach which will guide users through how to improve their lifestyle, we have identified the need to provide users with an estimation of their dementia risk relative to their age-matched peers. Hence we would be looking as well on the impact of age on the distribution of modifiable risk factors.