Approved research

Activity data as an input to EPSRC funded project ?The Wearable Clinic: Connecting Health, Self and Care?

Lay summary

We aim to enable new forms of collaborative care for long-term conditions by integrating data from wearable sensors (accelerometery) with the prediction of health risks from electronic health records. This will let us create algorithms for adaptive, personalized care planning that takes account of individual predicted risks and real-time sensing. We will investigate serious mental illness (schizophrenia), chronic kidney (renal) disease, and controls. We will quantify episodes of activity data as different measures/risk factors, use this as an input to our predictive modeling, and contributes toward an understanding of whether activity monitoring can be used an indicator of remission/relapse. We are developing software tools which will help patients with long-term conditions, together with their carers and doctors, to better manage their health in daily life, respond more quickly to changes, and prevent fall back episodes. By identifying associations between unsupervised behavioral phenotype (mobility, rhythmic/routines, sedentary behavior, fitness/frailty issues, weight gain/loss) and disease progression stages, we will build new disease risk prediction models. We have a strong focus on translation, with a dedicated 3 year work package on health and economic benefit analysis, so we can advance diagnosis, prevention/early detection, and risk stratification for chronic kidney disease and schizophrenia. The analysis stage will identify indicators (i.e. patterns/phenotype/risk factors) in the accelerometer data that assign patients to different disease states.(?Unsupervised? machine learning methods will be used to identify patterns in the data itself, without any intervention from the user, or a human interpreter.) These accelerometer based indicators will be used as one input, together with data on hospitalizations, self-reported measures, blood samples, and similar, for developing risk prediction models of how likely each person is to relapse, or to be in remission, or to be in another identifiable state which might alter their care planning.