A0766
Title: Generalized odds: Distributional object to model physical activity
Authors: Pratim Guha Niyogi - University of Mississippi Medical Center (United States) [presenting]
Kathryn Fitzgerald - Johns Hopkins University (United States)
Ellen Mowry - Johns Hopkins University (United States)
Vadim Zipunnikov - Johns Hopkins University, Bloomberg School of Public Health (United States)
Abstract: The advancement of mobile health (mHealth) and wearable technologies has provided valuable insights for medical and public health research, enabling a deeper understanding of the association between human behaviors and their impact on health. These data resources capture metrics such as physical activity (PA) and heart rate in near real-time, which have become increasingly popular due to their availability and wide range of applications in clinical research. Moreover, research on modeling the subject-specific distributional representations of such data has expanded considerably. The distribution representation of PA could be measured by the class of conditional and unconditional probability measures, such as survival and hazard functions, among many others. Generalized odds can provide a wide class of different distributional representations of the underlying data. This also produces a more nuanced interpretation of the relationships between PA and clinical outcomes. Motivated by the HEAL-MS study, a model that leverages scalar-on-distributional regression techniques is introduced to connect traditional measures of MS progression with this rich class of digital health data.
