Utilizing Wearable Devices to Improve Precision in Physical Activity Epidemiology: Sensors, Data and Analytic Methods

In physical activity (PA) epidemiology, precisely quantifying benefits of PA in terms of intensity, frequency and duration has been challenging, partly due to the difficulty and complexity in measurement. Traditional self-reported questionnaires are known to be subject to reporting bias and substantial measurement error. Enabled by technological advances in the past decades, wearable devices (including research-grade accelerometers, heart rate monitors and fitness trackers, etc.) have been increasingly adopted in large-scale epidemiological studies, recording massive amounts of high-resolution data. To take advantage of the rich data resources from these studies, advanced analytic tools are needed to rigorously quantify health benefits of patterns of PA and to inform effective intervention strategies. We first provided a review of a few recent developments in this area, covering topics on analytic methods for various aspects of accelerometry data processing and analysis, including standard analysis based on summarized data, alternative acceleration-based metrics from high resolution raw data and machine learning approaches for PA behavior classification. We then introduced a flexible functional data analysis approach for quantifying PA-health association across varying PA intensity levels. To showcase how the real world application of these methods, we applied them to the Objective Physical Activity And Cardiovascular Health (OPACH) Study to quantify does-response relationships between PA intensity and cardiometabolic risk factors among older women. We also discussed possible directions for future research.