STATISTICAL INNOVATIONS IN RISK MODELING

点击翻译按钮获取中文摘要

An overarching goal of this Project is to further understanding of breast, colon, and ovarian cancers, by developing and applying new methods for data analysis. Specifically, drawing on data from the cohort and from a new diet validation study to be conducted in 2006, we will address aims related to measurement error correction with repeated diet assessment. A second aim relates to the examination of combined endpoints. We will refine methods to apply polychotomous logistic regression to outcomes in Projects 1 to 3, we will apply methods to specific topics including: a. Risk factors for ductal vs. lobular breast cancer b. Dietary exposures (alcohol, folate, etc.) vs. ER/PR status in breast cancer c. Risk factors for ovarian cancers (i). Serous vs Mucinous vs. endometrioid / clear cell ovarian cancers (ii). Mucinous vs. non-mucinous ovarian cancers Risk factors for total mortality, including cardiovascular, cancer, and other endpoints, as well as leading cancer endpoints individually, will also be examined. Statistical models, which assume constancy of relative risks for a given risk factor, may be inappropriate when a combined outcome is being evaluated, though such models are often used. Using the methods of log-incidence modeling previously applied to breast and ovarian cancer, we will draw on data from Project 2 to develop a model for colon cancer. We will also evaluate causal inference in the context of breast cancer risk factors, and haplotype estimation and geneenvironment interactions. The close collaborations among investigators in this project with those in projects 1, 2, and 3, has led to the development and application of methods and important insights into cancer etiology. Continued collaborations with investigators addressing specific cancers will provide synergy in future investigations. The close work among the investigators from various projects maximizes efficiency, provides important input from a variety of sources, prevents duplication of effort, and results in a more coherent presentation of results across endpoints.