Multiple Imputation Based on Functional Principal Components Analysis for Sparse Longitudinal Data

• Activity Number: 106
• Type: Contributed
• Date/Time: Monday, July 30, 2007 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Szu-Ching Tseng*+ and Xiaowei Yang and Hao Zhang
• Companies: University of California, Davis and University of California, Davis and University of California, Davis
• Keywords: Sparse Longitudinal Data ; Missing Data ; Functional Principal Componenets Analysis ; Linear Mixed-Effects Models

Abstract

In biomedical research, due to unbalanced design with possibly large numbers of missing values, sparse longitudinal data are popularly seen in statistical practice. Recently, several solutions based on principal components analysis (PCA) for functional data have been developed. This paper aims to extend these methods within the framework of multiple imputation. By applying the PCA models to the sparse data with a chosen common time grid, complete data sets can then be generated. Each of these balanced data sets can be finally analyzed using standard longitudinal models such as random-effects models. By applying this imputation strategy to both simulated data sets and a hemodialysis vascular access data set, the performance of it is evaluated via comparing with the analysis using linear mixed-effects models to the original sparse data.