Statistical confirmation of negative results of association studies in genetic epidemiology

One of the most important reasons warranting the common reservation about publishing association studies with negative results is due to the fundamental fact that an insignificant result of a statistical testing procedure tailored for establishing an association, fails to provide conclusive evidence of the contrary. In this contribution, we show how the methodology of equivalence testing, as provided by modern biostatistics, can be exploited for removing this basic flaw. In order to keep the exposition as simple as possible, we restrict discussion to the setting of a study which aims at ruling out that some given single SNP has a relevant impact on the risk of developing the disease under consideration. As solutions to the problem of how to perform a valid confirmatory analysis of a trial of this type, two different approaches are presented: (i) exact Fisher type test for equivalence of an odds ratio to unity; (ii) distribution-free test for equivalence of cases and controls with respect to the full (i.e., trinomial) genotype distributions. The practical implementation of both testing procedures are described in detail and illustrated with examples taken from recent studies of the genetic epidemiology of psychiatric disorders.