Approved research

A retrospective investigation of accelerometer-derived movement patterns, with the aim of validating a machine learning algorithm, and potential identification of novel signs of disease

Lay summary

We aim to look at the movement data in UK Biobank data using a computer program that has been designed to spot patterns in large volumes of data. We will anonymously group people by medical history, age and gender (amongst other criteria), and observe for similarities within these groups. We will then tell the computer program what some of these represent, and it will learn to recognise changes from these patterns in the future - for example, the tremor observed in Parkinsonism and the limp seen in Osteoarthritis of the hip. It is possible that the program will discover similarities that have not been previously noticed by people. This might mean that we are able to discover new patterns of movement, that can be used to help diagnose people with illnesses in the future. We believe that this project will take 1-2 years to complete fully, as the development of the computer program will be dependent on the data that we receive (and so it is very difficult to predict the duration). This project presents the potential to help diagnose illness on a large scale by using the technological advancements (for instance, smart watches) that people already use. The opportunity to find new movement patterns is also an exciting one, and may be something that could help people even without access to wearable technology.