University of Cambridge > > Centre for Mobile, Wearable Systems and Augmented Intelligence Seminar Series > Can machine learning help identify new health relevant signals from large accelerometer datasets?

Can machine learning help identify new health relevant signals from large accelerometer datasets?

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Lorena Qendro.

Abstract: Cardiovascular disease (CVD) prevention strategies include the typical use of risk stratification and prediction tools to target preventive interventions for people at higher risk of CVD and recommendations on modifiable CVD risk factors such as levels of physical activity. However, most people with CVD are identified too late and over 50% of major cardiac events are in patients who were not classified as high-risk. In addition, a reliance on crude self-reported questionnaires could mean that CVD behavioural risk factors such as physical activity and sleep duration are more important than previously thought.

Wearable sensors such as wrist-worn activity trackers (accelerometers) have the potential to continuously, noninvasively, and painlessly measure CVD risk factors in patients’ everyday lives. For example, our group has worked closely with UK Biobank to measure physical activity status in 103,712 participants who agreed to wear a wrist-worn device for seven days. These measurements are now actively used by health researchers worldwide to demonstrate associations between physical activity and CVD . Machine learning methods can help maximise the utility of data from wearable sensors. However, there is a broad concern around the lack of reproducibility of machine learning models in health data science. It is critical to carefully consider how to promote robust machine learning findings and reject irreproducible ones, to ensure credibility and trustworthiness.

In this talk I will share my group’s work on reproducible machine learning of wearable sensor data for the early detection of cardiovascular disease. This will include methods to identify physical activity behaviours in a free-living validation dataset of 150 adults. I will then illustrate the genetic architecture of these measurements. I will also show that these measurements have a clear utility to predict future CVD outcomes. Finally, I will discuss the opportunities for wearable sensors to advance the prevention of CVD .

Zoom meeting:

Bio: I am an Associate Professor at the University of Oxford and lead Health Data Research UK’s national implementation project on reproducible machine learning. My research group at Oxford develops reproducible methods to analyse wearable sensor data in very large health studies to better understand the causes and consequences of disease. For example, we have developed methods to objectively measure physical activity in UK Biobank which are now actively used by researchers worldwide to demonstrate new associations with cardiovascular disease, depression, mood disorders, and others. We have also developed machine learning methods to identify sleep and functional physical activity behaviours such as walking. In addition, we have discovered the first genetic variants associated with machine-learned sensor phenotypes. This work shows the first genetic evidence that physical activity might causally lower blood pressure. In 2015 I was one of only three EU Marie Curie Award winners (from 9000 fellowship holders), selected for my contributions to health sensor data analysis. I have also contributed to the creation of guidelines on the use of mobile devices in clinical trials, in collaboration with the US Food and Drug Administration (FDA) supported Clinical Trials Transformation Initiative on “Mobile Clinical Trials”.

This talk is part of the Centre for Mobile, Wearable Systems and Augmented Intelligence Seminar Series series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2020, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity