A1897
Title: Inference and variable selection for two-phase studies with high-dimensional covariates
Authors: Haoyang Wang - The Hong Kong Polytechnic University (Hong Kong)
Qingning Zhou - University of North Carolina at Charlotte (United States)
Kin Yau Wong - Hong Kong Polytechnic University (Hong Kong) [presenting]
Abstract: The two-Phase study design is widely used to improve estimation efficiency and reduce cost. In many two-Phase studies, the outcome and inexpensive covariates are obtained on all subjects in Phase I, while expensive covariates are measured only on a subset of subjects in Phase II. As a result, regression analysis of two-Phase studies faces a missing data problem. When two-Phase studies involve high-dimensional covariates, a more challenging high-dimensional missing data problem arises. For this problem, complete-case analyses are generally inefficient, while imputation or likelihood-based methods require a model for the missing covariates, which is almost impossible to correctly specify. To overcome these limitations, a two-step estimation method is proposed that refines a complete-data estimator by incorporating the incomplete data and auxiliary information if available to improve efficiency. This method does not require modeling the distribution of the missing covariates, and the resulting estimator is guaranteed to be at least as efficient as the complete-data estimator. Theoretical properties of the proposed method are established, including estimation consistency, inference validity, and variable selection consistency. The performance of the proposed method is evaluated via simulation studies and an application to a major cancer study is provided.
