Single- and multi-locus association tests incorporating phenotype heterogeneity

Abstract

Taking disease subtypes into account when testing for an association between genetic factors and disease risk may help to identify specific aetiologic pathways. One way to assess a genetic association, whilst accounting for heterogeneity, is to use polytomous regression. This approach only allows heterogeneity to be considered in terms of a single categorical variable. In this article, we describe an alternative and novel test of association which incorporates multivariate measures of categorical and continuous heterogeneity. We describe both a single-SNP and a global multi-SNP test and use simulated data to demonstrate the power of the tests when genetic effects differ across disease subtypes. Applying the tests to the study of genetic variation in the oestrogen metabolic pathway and its association with breast cancer risk and prognosticators strengthened our understanding that the modulation of aromatase activity can influence the occurrence of tumours, and their grade and size, in postmenopausal women.