Study Design for Longitudinal and High Dimensional Measures.

Study Design;Longitudinal Measures;High-dimensional Measures;Classification;Optimal Design;Statistics and Numeric Data;Science;Biostatistics

Wu, MeihuaSong, Peter Xuekun ;

University of Michigan

关键词: Study Design;    Longitudinal Measures;    High-dimensional Measures;    Classification;    Optimal Design;    Statistics and Numeric Data;    Science;    Biostatistics;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/97810/meihuawu_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Study design is the foundation of successful clinical or epidemiologicalstudies. Ever since the seminal work of Fisher (1935), research in this area hasblossomed and many innovative concepts and approaches have been developed.Despite extensive literature on study design, new challenges for study designcontinue to emerge as innovative technologies push the limits of what can beinvestigated with a clinical or epidemiological study. For instance, tools forecological momentary assessment of behaviors or biological markers, or highthroughput experiment devices such as microarrays open the opportunity tomeasure complex biological processes over time, or the expression levels ofmillions of genetics or proteomics biomarkers simultaneously. In thisdissertation, we develop novel design methodologies for studies employing thesenew data collection techniques, namely: 1) studies involving repeated measuresof nonlinear profiles in biomarker studies with the objective of estimatingfeatures of the profile; 2) studies involving data with underlying functionalresponse with the objective of capturing the mean profile and between subjectvariability; 3) studies involving high dimensional genetics and proteomics datawith the objective of constructing classifiers with high probability of correctclassification. Correspondingly, our research is motivated by three practicalapplications: 1) salivary cortisol studies for investigating the associationbetween cardiovascular disease and stress; 2) urinary progesterone studies forreproductive health; 3) studies involvinghigh dimensional genetics and proteomics data with the objective ofconstructing classifiers with high probability of correct classification.This dissertation contributes novel study designmethodologies for studies that involve related but distinct data structures. Wedemonstrate the use of the methods with various examples to enhance thepotential of their used across a variety of settings. The new design methodologywill thus enable investigators to better evaluate the feasibility andcost-efficiency of the study in the planning stage and ultimately improve thechance of success of studies involving longitudinal and high dimensional data.