Search:

Cancer Control Research

Abstract

DESCRIPTION (provided by applicant): Exposure measurement errors in cancer epidemiology pose special methodologic challenges. For example, nutritional exposures form the basis for many etiologic hypotheses concerning cancer. However, nutrient intake is difficult to measure precisely. Other important examples considered in this grant include genomic risk factors and environmental exposures. It is the role of measurement error correction methods to estimate the relationship between cancer outcomes and exposures. To accomplish this requires both a main study where disease and the surrogate exposure are measured and validation data to determine the extent of the measurement error. In this proposal, we seek to continue our group's previous work on methods of corrections for measurement error and misclassification. A major focus is on nutritional studies based on intakes reported at a single survey when the target exposure is long-term average diet. Here, the dependent variable is time to cancer incidence or mortality; thus, Cox models are considered in which time-varying covariates such as cumulative averages and cumulative exposures are of primary interest. Another goal is to extend methods for random and subject-specific error terms when the usual exposure measurements and the reference exposure measurements may be correlated. These methods will be applied to a recent USDA study of assessment of energy intake using doubly labeled water and a re-analysis of the NCI's OPEN study of assessment of energy intake. Instrumental variables methods for nonlinear environmental exposure response estimation developed in the previous grant period will be extended to address gene expression response from DNA microarray data assessed longitudinally. These methods are to be applied to data from the New Hampshire Arsenic and Cancer Study. Finally, methods are proposed for genetic and genomic data increasingly being used in cancer epidemiology. Here, we propose to assess the extent to which bias in relative risk estimates is induced by SNP and haplotype misclassification, and develop new corrected estimators. User-friendly, public use software development will be a focus of all new methods development.