Survival data with missing time-independent or time-dependent covariates are commonly encountered in prognosis studies. Existing approaches mostly focus on the standard proportional hazards model, which may be too restricted in some applications. To obtain a comprehensive understanding of data, it is also important to identify variables that are associated with survival time. In this study, the investigators will develop nonparametric approaches for more flexible models with missing covariates, and investigate variable selection in this complicated setup.
The proposed research is expected to have broad impacts and application in biomedical studies, econometrics, environmental studies, behavioral and social sciences, where information is collected on time to an event of interest and multiple covariates. This proposed study will foster collaborations among investigators from different institutions/departments and backgrounds. It will promote teaching, training and learning in the Statistics Department and the newly founded Epidemiology and Biostatistics Department at the University of Georgia. Research conducted in this study will help develop advanced graduate courses in survival analysis, missing data and variable selection. It will create challenging statistical projects for graduate students that the investigators are supervising. Research results from this proposal will be disseminated through presentation at major statistical meetings. Software developed will be made publicly available, so that the proposed methods can be readily used in practice in various fields.