Understanding the association between risk factors and chronic diseases is crucially important for improvement in treatment and patient care. Population based data registries for chronic disease are often available and provide excellent platforms to serve this purpose. A crucial aspect of traditional regression modeling is the assumption that the effect of each risk factor is constant. However, there is growing evidence that the etiology of many chronic diseases is complex and the influence of a risk factor on disease outcomes may not be constant. Through our ongoing collaborative work on Cystic Fibrosis Foundation patient registry (CFFPR), the investigators of this grant have demonstrated that varying coefficient regression, termed as dynamic regression in survival analysis (Martinussen and Scheike, 2006), provides a powerful tool for discovering important changes in associations between risk factors and CF outcomes (time to, and frequency of, major CF events). This research project is motivated by several important unsolved, open questions arising from CFFPR: (i) competing risk, double censoring and left truncation to the observation of CF events, inherited with the design of CFFPR;(ii) the high dimensional covariates, which necessitate the development of variable selection procedures that accommodate varying covariate effects, (iii) the large sample size, which demands efficient computation;(iv) the need of utilizing time-dependent follow-up information to aid in disease prognosis and address substantive scientific questions. Current dynamic regression approaches have several limitations: inability to accommodate complex features of data, difficulties in interpretation and prediction, computational issues. Moreover, there is very limited work on variable selection under survival varying coefficient models. The overall objective of this proposal is to develop a comprehensive dynamic regression framework that resolves the key limitations of the existing approaches and possesses the capacity to handle many realistic data-related issues. To accomplish this goal, we will first lay out a unified framework of survival dynamic regression by introducing sensible modeling and developing inferential procedures that account for common survival data features (Aim 1). We will tackle the challenging problem of high dimensional dynamic regression (Aim 2), where the existing methods that assume constant effects can have poor performance. We will propose a seminal dynamic regression strategy for investigating the relationship between time-dependent covariates and survival outcomes with sensible interpretations and predictions permitted (Aim 3). The proposed statistical methods will be applied to CFFPR (Aim 4) and user-friendly software will be develop and made available to general research communities (Aim 5). Methodological development proposed in this grant will have a broad impact on scientific investigations not only on CFFPR but also on other registry based chronic disease studies.

Public Health Relevance

We propose statistical methods to identify risk factors for chronic disease studies, such as Cystic Fibrosis. These methods will enhance the understanding of the mechanism and prognosis of chronic diseases that will lead to improved disease treatment and patient care.

National Institute of Health (NIH)
National Heart, Lung, and Blood Institute (NHLBI)
Research Project (R01)
Project #
Application #
Study Section
Special Emphasis Panel (ZRG1-HDM-T (90))
Program Officer
Banks-Schlegel, Susan P
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Emory University
Biostatistics & Other Math Sci
Schools of Public Health
United States
Zip Code
Peng, Limin; Xu, Jinfeng; Kutner, Nancy (2014) Shrinkage Estimation of Varying Covariate Effects Based On Quantile Regression. Stat Comput 24:853-869
Ji, Shuang; Peng, Limin; Li, Ruosha et al. (2014) ANALYSIS OF DEPENDENTLY CENSORED DATA BASED ON QUANTILE REGRESSION. Stat Sin 24:1411-1432
Li, Ruosha; Peng, Limin (2014) Varying coefficient subdistribution regression for left-truncated semi-competing risks data. J Multivar Anal 131:65-78