Qingyuan Zhao1, Jingshu Wang1, Zhen Miao2, Nancy Zhang1, Sean Hennessy2, Dylan Small1, and Dan Rader2
1Department of Statistics, Wharton School, University of Pennsylvania
2Perelman School of Medicine, University of Pennsylvania
Earlier MR studies (Voight et al. 2012; Zhao, Chen, et al. 2019) found evidence of heterogeneity for the effect of HDL cholesterol on coronary artery disease (CAD). In this report, the research question we aim to explore is:
Do lipoprotein subfractions have heterogeneous effects on CAD?
To answer this question, we will conduct a MR screening study that estimates the causal effect of each subfraction on CAD. Some highlighting features of our analysis include:
Genome-wide three-sample design: This design allows us to utilize insturments that are only weakly correlated with the subfraction trait (may not be genome-wide significant). This is crucial because usually only a handful of genetic variants have genome-wide significant association with subfraction traits.
State-of-the-art statistical method: We used Robust Adjusted Profile Score (RAPS) which is not biased by individually weak instruments, as long as the overall instrument strength is not weak. RAPS is also robust to balanced and/or sparse pleiotropy; in particular, RAPS is asymptotically unbiased if the InSIDE (INstrument Strength Independent of Direct Effect) assumption is satisfied.
Multivariate MR: We applied a novel extension of RAPS to multivariate exposures. This allows us to assess whether the (potential) effect of any lipoprotein subfraction is independent of the major lipid traits (HDL-C, LDL-C, TG).