Intensive Longitudinal Data Models for Wearable Device Data (304904)*Jodi Lapidus, Oregon Health & Science University
Keywords: intensive longitudinal data, wearable devices, mixed models, personal health monitoring
The availability and use of wearable technologies -- body worn sensors that collect vital signs, activities, and other health indicators -- has exploded in recent years. The rich longitudinal data these devices produce has contributed to an increase in individuals' health awareness. Wearable device data are also routinely incorporated into observational cohort and intervention studies; however crude summary measures (avg minutes non-sedentary activity) may be analyzed rather than the entire data stream. Heterogeneous mixed effects regression models are an ideal strategy for wearable device data, as they can accommodate frequent and unequal measurements per subject, and can allow random effects to vary between groups of subjects.
Based on data from an occupational cohort of ~300 individuals, we examined how daily activity levels and changes in those levels influence resting heart rate variation over 4-6 months. We demonstrate how to parameterize fixed and random components to estimate activity-related associations with intra- and inter-individual heart rate variation. We show the usefulness of this strategy for elucidating complex associations in wearable device data.