Reducing sample sizes in genome scans: Group sequential study designs with futility stops

Authors

  • Inke R. König,

    1. Institute of Medical Biometry and Statistics, University Hospital Schleswig-Holstein, Campus Lübeck, University at Lübeck, Lübeck, Germany
    Search for more papers by this author
  • Helmut Schäfer,

    1. Institute of Medical Biometry and Epidemiology, Philipps-University of Marburg, Marburg, Germany
    Search for more papers by this author
  • Andreas Ziegler,

    Corresponding author
    1. Institute of Medical Biometry and Statistics, University Hospital Schleswig-Holstein, Campus Lübeck, University at Lübeck, Lübeck, Germany
    • Institute of Medical Biometry and Statistics, University Hospital Schleswig-Holstein, Campus Lübeck, University at Lübeck, Ratzeburger Allee 160, House 4, 23538 Lübeck, Germany
    Search for more papers by this author
  • Hans-Helge Müller

    1. Institute of Medical Biometry and Epidemiology, Philipps-University of Marburg, Marburg, Germany
    Search for more papers by this author

Abstract

Group sequential study designs can greatly facilitate analyses of genetic linkage in complex traits. We recently proposed designs allowing stopping investigations early if the result is significant (König et al. [2001] Am. J. Hum. Genet. 69:590–600), thereby decreasing average sample sizes under the alternative hypothesis. However, average sample sizes were slightly increased under the null hypothesis. We now present designs where the analysis of markers is additionally stopped in case of futility, i.e., if the probability for significant results is sufficiently low. These sequential designs are applied to linkage analyses of single loci. We calculated sample sizes, time points, and critical boundaries for all analyses for 2- and 3-stage designs at an overall significance level of 0.0001. To confirm the validity of asymptotic approximations, Monte Carlo simulations were performed. The utility is demonstrated analyzing genome scan data provided for the Genetic Analysis Workshop 12. Application of the novel sequential designs yields tremendous decreases in average sample sizes, regardless of the size of the underlying genetic effect at investigated loci. Depending on the applied design, almost half of the sample size is spared on average. These enormous savings are expected to have a special impact on costs and time of large-scale studies such as genome scans. Genet Epidemiol 25:339–349, 2003. © 2003 Wiley-Liss, Inc.

Ancillary