The Phenome-wide association study (
PheWAS) has become widely used for efficient, high-throughput evaluation of relationship between a genetic variation and a large collection of clinical diagnoses, typically extracted from a DNA biobank linked with electronic health records (EHR). The case-control analysis has been the standard choice for its efficiency in rare disease analysis and power to identify risk factors. However the clinical diagnoses in EHR are often inaccurate with errors and may cause bias. Here, we hypothesize that, case-cohort analysis, using relative risk (RR), is an alternative strategy that may overcome some of the limitations imposed by the case-control
PheWAS analysis. In the presentation, I will talk about the basic concepts for
PheWAS, the most common type of misclassification in
PheWAS and current solution to the issue. (Bias-corrected) Quasi-Poisson model, gold standard logistic regression and logistic regression, ignoring the misclassification, will be simulated to evaluate the bias in estimates under varying levels of misclassification. The pros and cons of all the models will be shown and discussed. Lastly, we will apply the models to the selected SNPs whose disease associations are well-known and compared with the case-control analysis.
MRBIII, Room 1220
17 April 2019
1:30pm