Statistical Methods on Emerging Medical Studies.

[摘要] Many areas of medical research can benefit from the application of new statistical methods. In this dissertation, we demonstrate the possibilities in some emerging medical studies. In chapter 2, we describe the design of longitudinal trials with dichotomized outcomes. Longitudinal studies are often analyzed using Generalized Estimating Equations (GEE) or Quadratic Inference Functions (QIF). We explore QIF-based analysis for study design: (i) we derive and compared sample size and power for QIF and GEE; (ii) we propose an optimal scheme of sample size determination to overcome the difficulty of unknown true correlation matrix; and (iii) we show that QIF based analysis become more efficient as the number of follow-up visits increases. We illustrate that without sacrificing power, the QIF design leads to sample size an study cost savings than the GEE analysis. In chapter 3, we focus on the analysis of exome sequencing studies. These studies focus sequencing resources on the exome but often yield many reads outside the targeted regions. These reads are almost always discarded but we propose our method SEQMIX to incorporate genotype likelihoods of off-targeted reads and show that it can decipher the local ancestry with resolution similar to that obtained with modern genome-wide association studies (GWAS) arrays. Our estimates are useful for analysis of human population history and disease gene mapping studies with exome and targeted sequence data.In chapter 4, we develop a general framework to explore benefits of incorporating genetic risk into prevention trial designs. We consider screening, recruitment and follow-up costs and current genetic findings in diseases. We consider type 1 diabetes (T1D), type 2 diabetes (T2D), myocardial infarction (MI) and age-related macular degeneration (AMD) as examples as they have distinct genetic architecture: many small effect markers are associated with T2D and MI; whilst loci with large effect size have been identified for T1D and AMD. We quantify the benefits by illustrating settings where reduction of trial cost or duration is up to 70% and settings where savings are modest. The benefits depend on the disease genetic architecture, but we also project that benefits will increase.

[发布机构] University of Michigan

[效力级别] Longitudinal Studies With Binary Outcomes

[关键词] Quadratice Inference Functions;Longitudinal Studies With Binary Outcomes;Design Clinical Trials With Genetic Information;Local Ancestry Inference in Admixed Samples;Next Generation Exome Sequencing Studies;Efficient Clinical Trials;Genetics;Public Health;Statistics and Numeric Data;Health Sciences;Science;Biostatistics