Article
Author(s):
Researchers developed a framework to pull high-resolution digital phenotypes from wearable devices and use them to predict the risk of cardiometabolic disease.
High-resolution digital phenotypes collected by wearable devices could be used to better predict cardiometabolic disease risk and improve tailored health management, a study published in the Journal of Medical Internet Research suggests.
Consumer-grade wearables such as smart watches and fitness trackers record heart rates, step counts, and other health data in normal day-to-day conditions. Recent research has also demonstrated summary statistics from such wearables have potential uses for the longitudinal monitoring of health and disease states.
“Unlike clean data from controlled experimental settings, real-world wearable recordings tend to be irregular, contain missing stretches, lack clean context annotations, and have variable lengths,” the study authors wrote. “As such, analyses based on the naive application of general purpose time series feature extraction methods may not have ecological validity.”
Because of these reasons, the authors hypothesized higher resolution physiological dynamics and phenotypes recorded by wearables may be applicable to modifiable and inherent markers of cardiometabolic disease risk.
To observe this, the authors used a framework to extract the high-resolution phenotype data from wearables and applied it to a multimodal data set, using machine learning to model nonlinear relationships and model comparisons to assess the predictive value of high-resolution phenotypes.
They discovered that these high-resolution physiological features had a higher predictive value compared with typical baselines for clinical markers of cardiometabolic disease risk.
Compared with baselines, the models that performed best using high-resolution features had a 17.9% better Brier score when based on age and gender, and a 7.36% better score when based on resting heart rate.
Heart rates in different activity states also contained different types of information, the authors found.
“Heart rate dynamics in sedentary states are most predictive of lipid abnormalities and obesity, whereas patterns in active states are most predictive of blood pressure abnormalities,” they said. “Moreover, in comparison with standard measures, higher resolution patterns in wearable heart rate recordings are better able to represent subtle physiological dynamics related to genomic risk for cardiometabolic disease.”
According to the authors, the higher resolution phenotypes resulted in an improvement of Brier scores between 11.9% and 22.0% for predicting genomic risk. Further, case studies showed there are links between high-resolution phenotypes and clinical events.
Based on these findings, the authors highlighted 2 potential applications of the developed framework.
First, the study revealed new relationships between high-resolution heart rate dynamics and cardiometabolic disease risk.
“These findings highlight the value addition of assessing physiology in free-living activity states (beyond controlled clinical settings) for disease risk monitoring and management,” the authors said.
Second, the results offer a new perspective on links between data collected by wearables and genetic predispositions in cardiometabolic diseases.
“As these associations did not appear to depend on the presence or absence of manifest clinical risk markers, we posit that high-resolution phenotypes from wearables may capture subtle subclinical physiological changes stemming from latent predispositions to disease,” they concluded.
Reference
Zhou W, Chan YE, Foo CS, et al. High-resolution digital phenotypes from consumer wearables and their applications in machine learning of cardiometabolic risk markers: cohort study. J Med Internet Res. Published online July 29, 2022. doi:10.2196/34669