Abstract:
|
Electronic Health Records (EHR) have emerged as a major source of data for clinical and health services research. Despite great potential, the complex and inconsistent nature of EHR data brings additional challenges for many clinical studies. A key challenge is the EHR data errors resulting from phenotype misclassification. Inappropriate handling of such errors may lead to reproducibility of findings across studies, which raises a fundamental concern about the value of these researching findings. In this talk, we will present novel methods that can conduct association analysis with well controlled Type I error and competing statistical power by accounting for complex differential misclassification mechanisms. We will also present several case studies of the proposed method, compared to existing methods.
|