A Novel Analysis of Accelerometry Data: Calibration and Flexible Association Modeling in an Epidemiological Study of Older Women

Abstract

Accelerometers have been widely deployed to objectively measure and monitor physical activity and sedentary behavior in large epidemiological studies. The traditional summary metric, known as counts, summarizes raw high-resolution acceleration signals for a pre-specified epoch length (e.g., 1 minute). However, its definition is proprietary to device manufacturers, making it difficult to compare studies that use different devices. Alternative summary metrics have been introduced in recent years. This master thesis conducted a novel analysis of accelerometry data based on the activity index, a novel and transparent way to summarize the high-dimensional accelerometry data, among older women within the Women's Health Initiative. We first built calibrating equations of activity index for estimating metabolic equivalents (METs) and derive cutpoints to classify epochs into distinct physical activity intensity categories. We then utilized single variable and isotemporal substitution models to investigate associations of one or more physical activity intensity category and health outcomes, such as cardio-metabolic risk factors in the Objective Physical Activity and Cardiovascular Health (OPACH) Study. Further, we adopted a newly developed functional data analysis framework to quantify the dose-response relationships between continuous physical activity intensity and cardio-metabolic risk factors in the OPACH Study.