Rethinking wearables data to maximize the value for health prediction
Wearable devices are everywhere, but their health tracking features still remain largely a novelty for many individuals. While some devices offer individual datapoints that can help people track certain aspects of their health – heart rate, cycle tracking, or O2 saturation – the market has thus far fallen a little short of creating a holistic, predictive portrait of a user’s wellbeing from their watch or smart ring.
That could be because developers aren’t focusing on all the right elements, suggests a pre-print paper from Apple, which has not yet been peer reviewed. The study, published on arVix in June, contends that a person’s behavioral data is often more predictive of healthcare status than straight-up stats generated by commonly available sensors.
Combining basic sensor data with more advanced calculations of sleep patterns, physical activity stats, and cardiac fitness are more impactful for potentially helping users make decisions about their health as compared to basic sensor data alone, the researchers say.
“Integrating behavioral data with raw sensor signals offers a holistic view of an individual’s health, enabling models to generalize across a wider range of detection problems,” the researchers said.
Metrics like step count, gait stability, mobility, and VO2 max (how much oxygen the body can use during exercise) “are sensitive to an individual’s behaviors, rather than being driven purely by physiology,” explains the paper. “These characteristics make behavioral data particularly promising for such health detection tasks. For example, mobility metrics that characterize walking gait and overall activity levels may be important behavioral factors to help detect a changing health state such as pregnancy.”
The challenge is that there aren’t a lot of these “scores” available yet, limiting their utility for broader examinations of a user’s health status. Apple hopes to change that by developing new models to analyze wearable data and create more usable scores that determine health status with a greater degree of accuracy and confidence.
To prove their theory, the team developed the Wearable Behavior Model (WBM), that can gauge health and even predict existing health conditions, such as diabetes, heart disease, or neuropathy, using next-level health scores. They also compared the results of the WBM with sensor-only calculations and a combination of the two methodologies.
Using Apple Watch and iPhone data from more than 160,000 participants, the team ran 27 different behavioral metrics, such as active energy, walking pace, heart rate variability, respiratory rate, and sleep duration, through the model.
When evaluated on 57 dimensions, WBM alone outperformed a sensor-data-based model in 18 of the 47 health prediction tasks related to a pre-existing health status, such as whether a person takes beta blockers or whether they have atrial fibrillation, and in all except one of the tasks related to current, dynamic states of health, such as being pregnant or having a respiratory infection.
But the really strong results came when both models were combined, bringing together basic sensor data with the more complex calculations of WBM. When both models were used simultaneously, they consistently outperformed each model individually, achieving extremely high accuracy rates for certain health statuses like pregnancy (92% accuracy).
The study indicates that taking a more complex and holistic approach to analyzing wearable data could bring these devices to the next level of utility for the millions of users who already own one or more gadgets.
By making better use of the data that’s being generated non-stop every moment, digital health experts, consumers, and healthcare providers may be able to gain additional insight into rising risks and make better decisions about future care plans.
Jennifer Bresnick is a journalist and freelance content creator with a decade of experience in the health IT industry. Her work has focused on leveraging innovative technology tools to create value, improve health equity, and achieve the promises of the learning health system. She can be reached at [email protected].