MERIT Award Recipient: Raymond J. Carroll, Ph.D.
Measurement Error, Nutrition and Breast/Colon Cancer
In our work, we propose to pursue three separate lines of work: (a) nutritional epidemiology; (b) genetic epidemiology with applications to nutrition; and (c) colon carcinogenesis experiments in nutrition. In (a) and (b), our work is done in conjunction with a number of researchers at the National Cancer Institute, and uses data from NCI studies.
In our work in nutritional epidemiology, we propose to develop statistical methods applicable to cohort studies of diet and cancer. Dietary studies typically use such instruments as food frequency questionnaires (FFQ), 24-hour recalls and diaries. In our work, we will investigate two problems:
- We will develop methods to understand the distribution of episodically-consumed foods (e.g., dark green vegetables) in a population, and then to develop methods to study the risk/benefits of these foods on the eventual development of cancers.
- We will attempt to improve the efficiency of dietary measurement through the use of borrowing strength across instruments. Currently, instruments are evaluated on a nutrient-by-nutrient basis, but we aim to show that a combined analysis of multiple nutrients will improve efficiency of measurement, and will also improve statistical power for detecting diet-cancer relationships.
Our work on genetic epidemiology is aimed at the question of gene-environment interactions. In many cases, it is reasonable to assume that the genetic status and environmental exposures are independent in the population, possibly after some form of stratification. Our goal is to use this assumption to improve efficiency of analysis in retrospective studies (population case-control studies and family-based matched studies). Our specific aims are:
- Develop a general semiparametric approach for population-based case-control studies, one that includes the possibility of (a) missing genetic information; (b) unphased haplotypes; and (c) environmental exposures such as dietary intake that are subject to errors of measurement. Our methods will be much more powerful than standard case-control analyses.
- In family-based matched case-control studies, including sib-pairs matching, we will develop new methods based on general conditioning arguments that exploit the gene-environment independence assumption to vastly improve statistical power.
Our colon carcinogenesis experiments use animal models, in which rats or mice are given different diets and exposed to carcinogens and/or radiation. The overall aim of the project is to understand how diet affects the early stages of colon carcinogenesis. Working with cancer biologists at Texas A&M, we now measure multiple markers on the same cells, e.g., apoptosis, p27, cell differentiation and cell proliferation. Our aim is to develop statistical methods within the framework of hierarchical functional data that can help understand the relationships among these markers, and whether these relationships depend upon diet.