Sholom Wacholder

NCI, DCEG, BB

PLCO (Learn more about this study)

2007-0035

Jun 12, 2007

Impact of control composition on genetic association studies

Abstract. If convenience controls are adequate, then whole-genome scans and other case-control studies of genetic factors will be far cheaper and easier than if control selection needs to meet the traditional epidemiologic standards (1-3). Indeed, the use of non-specific controls has been justified by assuming that the controls represent the genotype distribution of the source population from which cases arose, once ethnic origin is accounted for. Thus, in discussion of genetic association studies, far more attention has been paid to population stratification than to other potential sources of error, such as differential participation or nonresponse (4). The impact of using convenience controls on study findings is a controversial question (5), but has not been evaluated empirically or theoretically. We wish to use the data from the recently completed CGEMS whole-genome scan of prostate cancer, which characterized approximately 550,000 SNPs for case-control studies from ten centers, to assess the effects of control composition, apart from population stratification, on bias, type I and type II error, power and required sample sizes, when state-of-the-art statistical and genomic methods (6-8) are used to eliminate population stratification are applied. If statistical properties are acceptable in practice, the simplification and economy from convenience controls in genome scans might be acceptable, despite our methodologic concerns.