Missing or mismeasured regression variables are frequently encountered in the analysis of cancer research data. This problem is a major issue in dietary research, for example. The broad goal of this research is to develop statistical methods for data analysis in the presence of missing or mismeasured regression variables, and to apply these methods to epidemiologic and clinical studies. Specific areas of research include the following: 1) development of a nonparametric regression calibration method for Cox regression when validation data are available; 2) development of methods for Cox regression with reliability data; 3) development of methods for logistic regression analysis using conditional exposure mean for missing or mismeasured exposures; and 4) development of methods for characterizing associations between multiple variables. Asymptotic theory will be developed and simulation studies will be conducted. The methods will be applied to data from the Women's Health Initiative, a large disease prevention trial and observational study involving 164,500 women with the objective of evaluating the benefits and risks of dietary modification, hormone replacement therapy and supplementation with calcium and vitamin D on the overall health of postmenopausal women. The investigator states that the proposed methods will also accommodate two-stage designs which are becoming popular in epidemiologic research and particular in genetic epidemiology.
