New research from NDPH shows that combining machine-learning techniques with wrist-worn activity trackers could help individuals optimise their daily physical activity to reduce the risk of developing heart disease.
We all know that physical exercise is good for us, yet it remains unclear what the best combination of different activities is for optimum health. But since increasing numbers of us use wearable devices (such as Fitbits and smart watches) to monitor our physical activity, the data captured by these could, with individual’s consent, be a powerful resource for researchers to investigate this. Up to now, a key barrier to using this data has been the need to classify and label different behaviours (e.g. sitting, light activity, vigorous exercise): a highly time-consuming process if done manually.
In a new study, published today in The British Journal of Sports Medicine, NDPH researchers developed a machine-learning model to automatically classify activity data recorded by wearable devices. This was used to investigate how different movement patterns are associated with the risk of developing heart disease.
To develop the model, the researchers recruited 152 adults in Oxford who each wore a wrist-worn activity tracker and a wearable camera for 24 hours. The participants also kept an activity diary for the same period. Using the camera images and the diaries, the researchers labelled the data captured from the activity trackers as follows:
- Sleep;
- Sedentary behaviour, such as sitting working at a computer or watching television;
- Light physical activity behaviours, such as cooking;
- Moderate-to-vigorous physical activity behaviours, for instance walking the dog or cycling.
Based on this annotated dataset, the researchers developed an algorithm to automatically classify movement patterns. This was then applied to data collected from over 87,000 UK Biobank participants who had no previous history of heart disease. These had been sent a wrist-worn activity tracker to measure their movement patterns over a 7-day period in 2013 to 2015.
Over an average follow-up period of 6 years, 4,105 of the participants developed heart disease (including strokes and heart attacks). The results clearly showed that reallocating time from any behaviour to moderate-to-vigorous physical activity behaviours, or reallocating time from sedentary behaviour to any other behaviour, was associated with a significantly lower risk of heart disease. This was after taking into account other potential risk factors, including age, diet, sex, alcohol consumption and smoking.
For an average individual in this group, reallocating just 20 more minutes from a typical day to moderate-to-vigorous physical activity behaviours was associated with a 9% lower risk, while reallocating 1 hour per day to sedentary behaviour was associated with a 5% higher risk.
Lead researcher Rosemary Walmsley said ‘Our study shows that combining machine-learning with data captured from wearable devices in population-level studies can lead to valuable new health insights. Dedicating more time to one activity means we have less time to spend on others, so we specifically investigated reallocating time between different activities.’
Associate Professor Derrick Bennett, one of Rosemary’s supervisors, added: ‘This supports the current WHO Guidelines on Physical Activity and Sedentary Behaviour, which recommend at least 150-300 minutes of moderate-to-vigorous physical activity per week and replacing time spent sedentary with physical activity at any intensity.’
Co-supervisor Associate Professor Aiden Doherty said: ‘This huge study of wearables and cardiovascular disease represents an important step forward in our ability to understand how daily physical activity behaviours are related to cardiovascular health.