Recent studies have described the potential for ″collider bias″ to modify the magnitude of genotype-phenotype associations, however the extent to which this effect can induce a completely false-positive association remains unclear. In a sample of 142,630 individuals from the UK Biobank study, inclusion of height (a ″collider″) as a covariate induces biologically spurious, but genome-wide significant, associations between autosomal genetic variants and sex. These associations are non-significant in models unadjusted for height. Our study underpins the importance of causal inference modeling in the design and interpretation of genetic (and non-genetic) association studies.
A robust example of collider bias in a genetic association study