TITLE:  Structural Nonparametric Methods for Estimation, Prediction and Tracking with Longitudinal Data

ABSTRACT:

Longitudinal analysis has three important objectives in biomedical studies: (a) estimating the time-varying population-average and subject-specific covariate effects on the outcome process of interest; (b) predicting the future subject-specific outcome trajectories; (c) evaluating the tracking abilities of important risk factors and health outcomes. Because longitudinal data (which is often referred as functional data) consist repeatedly measured outcome and covariate processes over time, they can be used to accomplish the above three objectives simultaneously. Popular parametric methods for longitudinal analysis, such as the generalized mixed-effects models, are often too restrictive and unrealistic for real applications because of their modeling assumptions. On the other hand, nonparametric models without any structural assumptions could be computationally infeasible and difficult to interpret. We present in this talk some structural nonparametric methods to accomplish the above three objectives, namely estimation, prediction and tracking, based on a class of nonparametric mixed-effects models. Our methods, which use either local kernel-type smoothing or global smoothing via B-splines, have the appropriate model flexibility and computational feasibility, and are useful to answer many scientific questions which could not be properly addressed by parametric or unstructured nonparametric regression models. We demonstrate the application of our methods through a long-term epidemiological study of pediatric cardiovascular risk factors and a series of simulation studies. Asymptotic developments of our methods suggest that the convergence rates of our smoothing estimators depend on the number of subjects as well as the numbers of repeated measurements.

This is the joint work with Xin Tian (OBR/NHLBI)

Bio: Colin O. Wu is senior Mathematical Statistician at the National Heart, Lung and Blood Institute, National Institutes of Health (NHLBI/NIH). He is also Adjunct Professor at Georgetown University School of Medicine and Professorial Lecturer at The George Washington University. Dr. Wu received Ph.D. in statistics from the University of California, Berkeley. His past academic positions include the University of Michigan, the Johns Hopkins University, and Guest Lecturer at University of Maryland, College Park. His professional activities include serving as Guest Editor for Statistics in Medicine, Associate Editor for Biometrics, Data Monitoring Committee member for the United Sates Department of Veterans Affairs, and Sphygmomanometer Committee member for the Association for the Advancement of Medical Instrumentation (AAMI). He is Elected Member of the International Statistical Institute, and Fellow of the American Statistical Association. He has published over 130 peer reviewed articles in statistics, biostatistics and medical journals.