| The dissertation focuses on two problems identified from the ongoing Multiethnic Cohort (MEC) study, i.e. two-stage genotyping design and population stratification. Three chapters follow a general introduction to relevant concepts in genetic epidemiological studies. Chapter one and two contain two published papers on optimal two-stage genotyping design, using Bonferroni correction and false positive rate (FDR) correction, respectively, for the overall type I error rate. The significance of this part is that a procedure was designed to search for the most cost-effective two-stage studies based on the current availability of high-throughput and very high-throughput genotyping platforms (fixed SNP array) and without considering any constraints on total sample size or available resources. The results are to provide guidance for two-stage designs under a wide range of assumptions on the per-genotype cost ratio and total number of markers. Chapter three describes the performance of the generalized linear mixed model (GLMM) in controlling for population stratification with simulated structured case-control data. Compared to Genomic Control (GC) and the relatively new EigenSTRAT methods, GLMM does not offer much advantage. To assess the likely significance of population stratification in the planned or future MEC association scans, Chapter 3 also examines evidence of population stratification within four ethnic groups (African Americans, Japanese, Latinos and Whites) represented in the newly available MEC breast cancer data consisting of 1400 SNPs. The results indicate moderate structure exists in African Americans and Latinos and the GC approach produced acceptable empirical type I error in most cases. |
