3.Digital phenotyping to demonstrate the impact of circadian aging using wearable data

The application of machine learning to continuously collected data from wearables elucidates hidden patterns as digital phenotypes and facilitates subpopulation identification. Conventional, expert-driven classification of disease or at-risk populations is limited by a lack of agreed ways of knowing the number of natural clusters in the populations of interest and determining the variables on which to base segmentation. Instead, the use of a holistic and data-driven clustering approach has gained recognition as an alternative. That is, each individual exists within multiple classes of health levels and provides various modalities of digitally measured physiological and behavioral data, which then correspond to multiple clusters of health status. To summarize, digital biomarkers and data-driven clustering approaches enable the use of precision medicine. These methods can classify a population into groups with unique characteristics or health risks and help individuals move from “unhealthy or at-risk” classes to “healthy” classes through intervention.

To this end, we investigate digital phenotyping as a means to demonstrate the effects of circadian aging derived from wearable data. Specifically, we applied data-driven clustering nad propensity score matching to identify distinct patterns in circadian rest-activity rhythms and the association with biological aging acceleration.