PLOS Genetics paper derives bias correction for liability-scale GWAS of covariate-adjusted binary traits
A new statistical framework addresses collider bias introduced when heritable covariates are adjusted for in binary GWAS, and provides the correct liability-scale conversion factor.
A methods paper published in PLOS Genetics by Siyi Chen derives a correction for a subtle but consequential form of bias that arises in genome-wide association studies (GWAS) of binary disease outcomes when heritable covariates — such as body mass index or a comorbidity — are included as adjustors in the regression model.
Adjusting for heritable covariates can induce collider bias, a form of statistical confounding that creates spurious genetic associations by conditioning on a common downstream effect of both the exposure and the outcome. The paper characterises this problem formally under a Gaussian liability model and derives the appropriate conversion factor for re-expressing log-odds effect sizes on the liability scale, which represents continuous, unobserved disease susceptibility. Crucially, the derivation shows that both the direct effect (excluding pathways through the adjusted covariate) and the collider-bias component share a common conversion factor, which simplifies correction.
The work is relevant to researchers conducting or interpreting binary-trait GWAS, particularly those in which covariate adjustment is necessary for biological or epidemiological reasons. Correct liability-scale estimates are important for downstream analyses including heritability partitioning, polygenic score construction, and genetic correlation estimation. The paper is published in PLOS Genetics and is available open access.
Sources
Read the original reporting — these are the public sources this summary draws from.
-
Primary source PLOS Genetics · 2026-10-30Conditional-prevalence correction for liability-scale analyses of covariate-adjusted binary GWAS