Grant: Discovery Grants Program - Individual

The main objective of this research program is to provide statistical methods and supporting theory for the study of large, complex datasets. For example, these datasets can result from national and international population cohort studies collecting rich information on biological and environmental factors, and complex traits and event outcomes. The need for new statistical methodology will continue to grow as science, measurement and information technology advance, yielding increasingly complex data. Two main areas of research are proposed: (a) methods for two-phase studies from large cohorts, and (b) methods and computational tools for analyzing complex mechanisms of association in large datasets. The specific projects will address a number of important gaps in the current statistical methodology that are limiting informative scientific analysis in many fields. Two-phase study designs are important for the analysis of large prospective cohorts, whereby certain variables (e.g. outcomes and covariates) are measured for all individuals (phase 1), and then additional (usually more expensive) variables of interest are measured for a subset of these individuals (phase 2). With the increasing wealth of data available from large, complex cohort studies, the analysis of two-phase studies is receiving considerable attention. A number of new important problems have presented themselves that require the development of new statistical methods. I will address three key topics related to (a): (i) methods for design and analysis of two-phase studies based on stratified sampling with complex covariate structure, (ii) multivariate analysis methods for studies involving many covariates of interest from large databases, and (iii) methods related to risk prediction from two-phase study data. To harness the full potential of new technological advances and integrate large datasets to capture complex mechanisms of association, the research community requires novel analytic tools. Topics related to (b) will focus on the development of suitable modeling and testing approaches for the effects of covariates on the variance (or scale) of count data, and new methodology for the analysis of data with sparse, complex covariate structure. This work will result in better study designs and tools for improved analysis of two-phase studies from large cohorts. The development of methods and computational tools for analyzing large and complex datasets will have implications to the broader research community where big data is emerging. Collaborations with researchers in areas of molecular biology, genetics, and evolutionary ecology will further ensure the relevance and impact of the proposed research program.