Statistical methods for analyzing objectively measured physical activity data

点击翻译按钮获取中文摘要

Project Summary/Abstract The major focus of this project is to develop novel statistical methods for objectively measured physical activity (PA) data by accelerometers or other wearable monitors, and to apply these methods to understand how specific patterns of activity intensity, frequency and duration are associated with health outcomes, such as cardiovascu- lar diseases. With rapid technological advances, it is now increasingly common to record accelerometer data in large-scale epidemiological studies. While the size and complexity of this type of data is exploding, the de- velopment of analytic methods is largely lacking. Novel analytic methods that are interpretable, robust, and yet computationally efficient are urgently needed, especially with the technology rapidly evolving. Motivated by the Women's Health Initiative (WHI) and other PA studies, we propose novel statistical meth- ods to summarize and effectively utilize objective physical activity measurements for health association analysis. Specifically, we will first develop a functional data analysis framework to extract novel summary indices from standard resolution accelerometer data and flexibly model the association between PA patterns and health. We next consider high-resolution raw accelerometer data and propose a systematic analytical framework for identifying walking in free-living environment, describing walking patterns and studying their association with health outcomes. Finally, we propose novel measurement error correction approaches for large-scale studies with complex sampling designs, where objective measurement is available only in a small subset of the study population. The proposed methodological research is motivated by PA studies of the WHI and will be directly applied to these projects. Applications of these methods to these studies will improve our understanding of PA, which will eventually help lay the foundation for future PA guidelines. These methods are generally applicable to other epidemiological studies that collect objective PA measurements, and software will be made publicly accessible.