Abstract
Genetic epidemiology studies often adjust for numerous potential confounders, yet the influences of confounder misclassification and selection bias are rarely considered. We used simulated data to evaluate the effect of confounder misclassification and selection bias in a case-control study of incident myocardial infarction. We show that putative confounders traditionally included in genetic association studies do not alter effect estimates, even when excessive levels of misclassification are incorporated. Conversely, selection bias resulting from covariates affected by the single-nucleotide polymorphism of interest can bias effect estimates upward or downward. These results support careful consideration of how well a study population represents the target population because selection bias may result even when associations are modest.
Highlights
Genetic epidemiology studies often adjust for numerous covariates when estimating associations between genetic variants and an outcome of interest [1,2,3]
We initially investigated whether acquired risk factors confounded the association between two simulated causal single-nucleotide polymorphisms (SNPs) and the odds of myocardial infarction (MI) using the Genetic Analysis Workshop 16 simulated data
To evaluate the effect of misclassification related to the true confounder values, we considered models that fixed the probability of smoking misclassification for non-smokers and Rx at 2% and varied the misclassification probability for smokers from 2-30%
Summary
Genetic epidemiology studies often adjust for numerous covariates when estimating associations between genetic variants and an outcome of interest [1,2,3]. Several authors have questioned whether acquired risk factors can confound genetic associations [4,5,6]. Others have suggested that confounders only influence genetic associations through selection bias [7,8]. We initially investigated whether acquired risk factors confounded the association between two simulated causal single-nucleotide polymorphisms (SNPs) and the odds of myocardial infarction (MI) using the Genetic Analysis Workshop 16 simulated data. This platform facilitated an evaluation of non-differential and differential misclassification in potential confounders, as well as selection bias
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have