Complex Survey-Weighted Functional Regression Methods for Mortality Prediction in National Health and Nutrition Examination Studies (NHANES)

• Activity Number: 104 - Novel Methodology for the Analysis of Physical Activity Data Measured by Accelerometers
• Type: Invited
• Date/Time: Monday, August 9, 2021 : 1:30 PM to 3:20 PM
• Sponsor: Biometrics Section
• Author(s): Ekaterina Smirnova* and Andrew Leroux and Erjia Cui and Lucia Tabacu and Ciprian Crainiceanu
• Companies: Virginia Commonwealth University and University of Colorado Anschutz Medical Campus and Johns Hopkins Bloomberg School of Public Health and Old Dominion University and Johns Hopkins University
• Keywords: NHANES; Functional Data; Accelerometry

Abstract

Understanding the interactions between physical activity (PA) and health is crucial for public health. PA data used in public health research is often collected using self-reports, which are subject to recollection and social desirability biases. To address this problem, a growing number of studies use accelerometers to objectively quantify physical activity in the free-living environment. Currently, the National Health Examination Study (NHANES) is the largest US-population study that contains publicly available PA data obtained from wearable accelerometers. Analyzing the association between PA in NHANES and mortality raises important and practical methodological challenges: (1) data is high dimensional with one observation per minute for up to seven days for each study participant; (2) NHANES study participants are recruited form the US population according to a survey design procedure; and (3) the structure of the association between PA trajectories and outcomes is a-priori unknown. I will describe the problem of predicting health outcomes (five-year all-cause mortality) using high dimensional data while accounting for survey design and weights.