Highlights:
- A study of over 100,000 adults found that motion sensor data from just 6 minutes of walking was enough to predict five-year mortality risk as accurately as other leading methods.
- The study used data from the UK Biobank study, which included participants wearing motion sensors on their wrists for one week.
- The researchers developed an algorithm that estimated mortality risk using acceleration data during 6- minute walk
- The study’s findings suggest that passive measures with motion sensors can be used to predict mortality risk and identify individuals who may benefit from interventions to improve their physical activity levels and overall health.
Data collected from smartphones and other wearable devices could be used to estimate an individual’s risk of dying in the next five to six years, according to a new study published in the journal PLOS DIGITAL HEALTH. The study conducted by the researchers at the University of Illinois Urbana–Champaign found that data from just six minutes of walking, collected via wearable motion sensors devices such as Fitness watches can predict an individual’s risk of death from any cause with up to 70 % accuracy.
Most people have smartphones with similar sensors, but calculating mortality risk from the data they collect is difficult because people do not usually carry their phones all day, says Bruce Schatz, author of the study and a university member. To find a measurable alternative to cell phones, Schatz and his colleagues analyzed data from 100,655 participants in the UK Biobank study, which collects information on the health of middle-aged and older adults who have lived in the UK for over 15 years. Participants in that study wore wrist devices with accelerometers for a week. Over the course of the next five years, about 2% of the participants passed away and this data was used for the study. The UK Biobank cohort is demographically representative of the UK population and the dataset represents the largest available sensor record of its kind.
The researchers ran motion sensor and death data on about one-tenth of participants through a machine learning model, which developed an algorithm that estimated five-year mortality risk using acceleration during a 6-minute walk.
They then tested the model using data from the other participants and determined its c-index score – a metric commonly used in biostatistics to assess accuracy – was 0.72, which is comparable to other metrics of estimating life expectancy, like daily physical activity or health risk questionnaires.
“For many diseases, specifically heart or lung diseases, there is a very characteristic pattern in which people decelerate when they are out of breath and accelerate again in short doses,” explains Schatz.
Although this study used wrist motion sensors, smartphones are also capable of measuring acceleration during short walks, the author adds, who is currently planning a larger study using smartphones. “If people carry phones, you can make a weekly or daily prediction and that is something you can’t get by any other method.”
While measurements of speed and heart rate by pulse are called active measurements, those made with smartphones, which can measure the intensity of walking while being in the person’s pocket, are called passive measurements.
“Our results show that passive measurements with motion sensors can achieve similar accuracy to active measurements of walking speed and walking pace,” the authors of the study said.
This is significant because it shows that the use of smartphone sensors for predictive health monitoring is possible and could potentially be used for large-scale population screening. This means that smartphones and other wearable devices could be used to predict an individual’s risk of dying, and provide targeted interventions to improve their health.
The researchers conclude that their scalable methods offer a feasible pathway toward national screening for health risks and could potentially revolutionize the way healthcare infrastructure is implemented.
The study used data from the UK Biobank, which collected questionnaires and laboratory test results from participants between 2006 and 2010. The researchers extracted 20 questions to characterize health status, including seven advanced disease conditions, 10 modifiable risk factors, and three demographics. The study also used accelerometer data collected with the Axivity AX3 wrist-worn accelerometer, which collected 100Hz triaxial signals. The researchers extracted 76-dimensional feature vectors from the raw data and used these to develop predictive models of mortality risk. The total processing time for feature extraction was 3100 hours of compute time, and the extracted features were 1.2 TB in size.
Previous studies have estimated mortality risk using daily physical activity levels tracked by the motion sensors, for example, a study conducted on more than 44 000 middle-aged and older individuals found that individuals who were physically inactive had a higher risk of dying from any cause, compared to those who were moderately or highly active. The study used data from wearable motion sensors to measure the physical activity level of more than 120,000 adults over a period of six years and found that those who were physically inactive had a 20-30% higher risk of dying from any cause compared to those who were moderately active.
The World Health Organization recommends that adults engage in at least 150 minutes of moderate-intensity physical activity or 75 minutes of vigorous-intensity physical activity each week. This can include activities such as walking, running, cycling, swimming, and other forms of exercise.
