Genome-wide association scans for prostate and breast cancer: design and analysis

P Kraft(1), D Hunter(1), S Chanock(2), R Hoover(3), D Gerhard(4), G Thomas(5)

(1) Prog. in Mol. and Genet. Epi., Harvard SPH
(2) Core Genotyping Facility, NCI
(3) Div. of Cancer Epi. and Genet., NCI
(4) Office of Cancer Genomics, NCI
(5) INSERM U434, Fondation Jean Dausset-CEPH

We discuss several problems in the design and analysis of genome-wide association 
scans using the Cancer Genetic Markers of Susceptibility (CGEMS) study as an 
illustration. CGEMS has adopted a multi-stage approach to scan the genome for 
prostate and breast cancer susceptibility loci. Initial stages involve nested 
case-control samples from ongoing cohort studies that have been enriched for 
advanced disease; over 5,000 cases of each cancer are available. We present a 
strategy for selecting SNPs to be genotyped at each stage and corresponding 
multi-stage power calculations, adjusting for the fact that susceptibility loci 
are unlikely to be directly observed but can be detected due to linkage 
disequilibrium (LD) with genotyped SNPs. In particular, we examine whether 
follow-up scans should have greater "breadth" or "depth": whether many promising 
SNPs should be tested without adding nearby SNPs to improve LD with unobserved SNPs 
or whether the few most promising SNPs should be comprehensively tested by 
genotyping more nearby SNPs. We suggest that a "broad" approach is appropriate for 
early stages where thousands of SNPs will be typed on a limited number of subjects. 
We also present a simple analytic strategy (based on 3?3 contingency tables) that 
retains power whether a locus broadly influences cancer risk or only influences 
risk of advanced disease, highlighting the potential gain from oversampling advanced 
cases.