Reliability of statistical associations between genes and disease

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 x 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 x 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.