A0907
Title: Multivariate principal component analysis for mixed-type functional data with application to mHealth in mood disorders
Authors: Debangan Dey - Texas A&M University (United States) [presenting]
Abstract: Mobile health studies collect various self-reported assessments capturing participants' behavior and well-being throughout the day. These assessments cover different scales, including continuous for physical activity, truncated for pain levels, ordinal for mood states, and binary for daily life events. Indexing these assessments by time and stacking them together, they form multivariate functional data with continuous, truncated, ordinal, and binary variables. A multivariate functional principal component analysis is proposed using a semiparametric Gaussian copula model, assuming a generalized latent non-paranormal process as the underlying mechanism. Latent temporal and inter-variable dependence is estimated through Kendall's Tau bridging method. The approach facilitates consistent function-on-function regression models, exemplified using data from 310 participants in the National Institute of Mental Health Family Study of the Mood Disorder Spectrum. The method characterizes the association between objectively collected physical activity and self-reported mood in individuals with major mood disorder subtypes, including Major Depressive Disorder and Type 1 and 2 Bipolar Disorder.
