Nonparametric Estimation of Time-Lagged Varying-Coefficient Models with Longitudinal Data

• Activity Number: 665 - Regression Methods for Longitudinal Data
• Type: Contributed
• Date/Time: Thursday, August 1, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Xin Tian* and Colin O. Wu and Xiaoying Yang and Zhaohai Li
• Companies: National Heart, Lung and Blood Institute, National Institutes of Health and National Heart, Lung and Blood Institute, National Institutes of Health and The George Washington University and The George Washington University
• Keywords: Longitudinal data; Varying-coefficient model; Time-lagging effect; Coefficient curves; Lagging time; Spline-based estimation

Abstract

Longitudinal data, where the subjects are repeatedly observed over time, have a unique capability of investigating the changes in disease risks and time-varying associations between risk exposures and health outcomes. By incorporating parametric modelling structures and flexible nonparametric coefficient curves, the time-varying coefficient models are a class of natural tools that describe the dynamic relationships of longitudinal variables. However, conventional time-varying coefficient models, which assume that the covariates affect the outcome instantaneously, are unable to evaluate the time-lagged covariate effects. We propose a class of time-lagging varying-coefficient models, which extend the conventional models to incorporate the covariate exposure over an unknown lagged time, and develop a spline-based procedure to estimate the time-lagging coefficient curves and select the appropriate lagging time. We demonstrate the performance of our proposed model through a simulation study and an application to the longitudinal biomarker data from a clinical trial of hematopoietic stem cell transplantation.