This project will develop and investigate new methodology for analyzing data from clinical studies of cancer and other chronic diseases. The specific areas of concentration are as follows. l. Methodology for Analysis with Missing Data. In clinical studies it is common to have cases with incomplete data, both on covariates and on outcome measures. Standard complete case methods of analysis applied to such data are often not valid. Several specific areas will be investigated: (a) Joint estimation of the covariate distribution and the regression model in a full likelihood analysis with missing covariate data, with an emphasis on """"""""nonpararametric"""""""" models for the covariate distribution. (b) Extending methods for handling missing covariate data from model fitting to methods for exploratory analysis and data smoothing. (c) Examining parametric methods for inference with non-ignorable missing outcome data. 2. Time- Varying Covariates. Clinical studies with failure time endpoints often include collecting data on marker process which vary over time. Methods for exploratory analysis through nonparametric estimation of the relationship between such timevarying covariates and failure time will be developed. 3. Transformed Linear Survival Models. Transformed linear survival models provide a useful alternative to the proportional hazards model for regression analysis of failure time data. New methods will be developed for estimating optimal weight functions for the rank based estimating equations for these models. New methods for flexible estimation of regression functions including generalized additive modeling, will also be developed.
Showing the most recent 10 out of 15 publications
