VARIABLE SELECTION FOR CASE-COHORT STUDIES WITH FAILURE TIME OUTCOME Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 19, 2019
Creator
  • Ni, Ai
    • Affiliation: Gillings School of Global Public Health, Department of Biostatistics
Abstract
  • Case-cohort design is widely used in large cohort studies with failure time data to reduce the cost associated with covariate measurement. Many of those studies collect a large number of covariates. Therefore, an efficient variable selection method is needed for the case-cohort design. In this dissertation, we study the properties of the Smoothly Clipped Absolute Deviation (SCAD) penalty based variable selection procedure in Cox proportional hazards model and additive hazards model in a case-cohort design with a diverging number of parameters. We prove that the SCAD penalized variable selection procedure can identify the true model with probability tending to one under Cox proportional hazards model. We then establish the consistency and asymptotic normality of the penalized estimator. We show via simulation that the BIC-based tuning parameter selection method outperforms the AIC-based method under typical case-cohort study settings. The proposed procedure is applied to the Busselton Health Study (Cullen1972, knuimanserum2003). Additive hazards model is a useful alternative to the Cox model for analyzing failure time data. In the second part of the dissertation, we extend the SCAD-penalized variable selection procedure to the additive hazards model with a stratified case-cohort design and a diverging number of parameters. We again establish variable selection consistency, estimation consistency, and asymptotic normality of the penalized estimator under this setting. We propose a new tuning parameter selection method and evaluate its performance via simulation. We show that the proposed tuning parameter selection method outperforms the conventional k-fold cross-validation method. The proposed procedure is applied to the Atherosclerosis Risk in Communities (ARIC) study (ARIC2004). Tuning parameter selection is critical to the success of a regularized variable selection method. A consistent tuning parameter selection method has not been established for the SCAD-penalized Cox model with a diverging dimension. In the last part of the dissertation, we propose a generalized information criterion (GIC) for tuning parameter selection and establish conditions required for its variable selection consistency under this setting. Simulation study shows that GIC performs well under the required conditions with finite sample size. It is then applied to the Framingham Heart Study (Framingham).
Date of publication
Keyword
Subject
DOI
Identifier
Resource type
Rights statement
  • In Copyright
Advisor
  • Cai, Jianwen
  • Zeng, Donglin
  • Bensen, Jeannette
  • Sun, Wei
  • Herring, Amy
Degree
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2015
Language
Publisher
Place of publication
  • Chapel Hill, NC
Access
  • There are no restrictions to this item.
Parents:

This work has no parents.

Items