Skip to navigation Skip to main content Skip to footer

Approved research

Predicting Risk for Chronic Conditions using Machine Learning

Principal Investigator: Dr Nils Hammerla
Approved Research ID: 26638
Approval date: June 30th 2017

Lay summary

Disease risk prediction relies on well established analysis tools and simple statistical techniques. More recent methods, such as machine learning, rely on large representative datasets that have previously been unavailable to validate such predictive modelling approaches. In this project we want to investigate if additional information related to an individuals' lifestyle can improve the reliability of disease risk prediction models. We want to investigate if recently developed methods for time-series analysis improve upon the state-of-the-art models for common chronic conditions such as diabetes, cardiovascular disease, and mood disorders. Our project is in line with the purpose of the UK biobank as we will use the resource to establish (freely accessible) disease risk prediction capabilities for our existing user-base in the UK, Ireland, and Rwanda. Any findings relating to current risk prediction models from the literature will be contributed back to the resource, such that future incident rates can be compared with gold standard estimates. Further we will do our utmost to publish clinically relevant findings, promoting the use of the UK biobank as a resource for similar future research projects. The data will be stored on our secure server which only named applicants will have access to. We will validate some of the published disease prediction algorithms using the UK Biobank data which has secondary linkage to longitudinal disease outcomes to assess their accuracy and performance. We will develop novel disease prediction algorithms using advanced machine learning techniques to be more accurate, applicable and scalable. Results will be disseminated in an academic journal and will be accessible through our free-to-use mobile app. For this project we would require the full cohort.