Title: Multilevel variance components model in minute-level accelerometry measures for twin studies
Authors: Haochang Shou - University of Pennsylvania (United States) [presenting]
Abstract: The emergence of mobile technologies, such as physical activity assessed via wearable actigraphy devices has provided an unprecedented opportunity to obtain objective evaluations of multiple physiological systems in real-time over weeks or months. However, the complexity of the devices and the high-dimensionality of the data also pose many analytic challenges to time-dependent measures. Most of the current approaches are based on summary statistics of activity that neglect the important time effects. We developed multilevel functional data analysis approaches that integrate multiple domains of complex measurements and reduce the dimensionality of the data while accounting for correlations in the repeated observations. In particular, motivated by the physical activity data observed from Brisbane adolescent twin study, we extended the traditional ACE model for a single univariate trait to functional outcomes based on an earlier work of structural functional principal component analysis (SFPCA). The method simultaneously: 1) handle various levels of correlation in the data; 2) identify interpretable traits via dimensionality reduction based on principal components; and 3) estimate relative variances that are attributed by additive genetic, shared environmental and unique environmental effects. Within-family similarities of those complex measures could also be effectively quantified.