PheWAS-based clustering of Mendelian Randomisation instruments reveals distinct mechanism-specific causal effects between obesity and educational attainment

Abstract

AbstractMendelian Randomisation (MR) is a statistical method that estimates causal effects between risk factors and common complex diseases using genetic instruments. Heritable confounders, pleiotropy and heterogeneous causal effects violate MR assumptions and can lead to biases. To tackle these, we propose an approach employing a PheWAS-based clustering of the MR instruments (PWC-MR). We apply this method to revisit the surprisingly large apparent causal effect of body mass index (BMI) on educational attainment (EDU):α= -0.19 [-0.22, -0.16].As a first step of PWC-MR, we clustered 324 BMI-associated genetic instruments based on their association profile across 407 traits in the UK Biobank, which yielded six distinct groups. The subsequent cluster-specific MR revealed heterogeneous causal effect estimates on EDU. A cluster strongly enriched for traits related to socio-economic position yielded the largest BMI-on-EDU causal effect estimate (α= -0.49 [-0.56, -0.42]) whereas a cluster enriched for primary impact on body-mass had the smallest estimate (α= -0.09 [-0.13, - 0.05]). Several follow-up analyses confirmed these findings: (i) within-sibling MR results (α= -0.05 [-0.09, -0.01]); (ii) MR for childhood BMI on EDU (α= -0.03 [-0.06, -0.002]); (iii) step-wise multivariable MR (MVMR) (α= -0.06 [-0.09, -0.04]) where time spent watching television and past tobacco smoking (two proxies for potential confounders) were jointly modelled.Through a detailed examination of the BMI-EDU causal relationship we demonstrated the utility of our PWC-MR approach in revealing distinct pleiotropic pathways and confounder mechanisms.