One-stage design is empirically more powerful than two-stage design for family-based genome-wide association studies

Finding a genetic marker associated with a trait is a classic problem in human genetics. Recently, two-stage approaches have gained popularity in marker-trait association studies, in part because researchers hope to reduce the multiple testing problem by testing fewer markers in the final stage. We compared one two-stage family-based approach to an analogous single-stage method, calculating the empirical type I error rates and power for both methods using fully simulated data sets modeled on nuclear families with rheumatoid arthritis, and data sets of real single-nucleotide polymorphism genotypes from Centre d'Etude du Polymorphisme Humain pedigrees with simulated traits. In these analyses performed in the absence of population stratification, the single-stage method was consistently more powerful than the two-stage method for a given type I error rate. To explore the sources of this difference, we performed a case study comparing the individual steps of two-stage designs, the two-stage design itself, and the analogous one-stage design.