Immunogenetics

, Volume 57, Issue 8, pp 549–558

Reliability of statistical associations between genes and disease

Authors

    • Pathology and Laboratory Medicine, Anatomy and Neurobiology, Center of Genomics and BioinformaticsUniversity of Tennessee Health Science Center
    • BiostatisticsUniversity at Buffalo
Review

DOI: 10.1007/s00251-005-0025-x

Cite this article as:
Manly, K.F. Immunogenetics (2005) 57: 549. doi:10.1007/s00251-005-0025-x

Abstract

Many statistical associations between a disease and alleles of specific genes have proven to be irreproducible. In part, this irreproducibility can be attributed to a lack of replication before publication and the fact that, until recently, the relationship between statistical significance and various measures of reproducibility was not widely understood. This review proposes a classification system, the Better Associations for Disease and GEnes (BADGE) system, for describing genetic associations. The BADGE classes, first class through fifth class, are based on the P value of the association. A first-class association, with P<2×10−7, is expected to be reproducible even in the absence of other evidence supporting the association. A fifth-class association corresponds to conventional statistical significance (P<5×10−2), which provides almost no assurance of reproducibility. Three intervening classes, described as second-, third-, and fourth-class associations, are defined by P values separated by factors of 20 or 25 from these extremes.

Copyright information

© Springer-Verlag 2005