PheBC: bias correction methods for EHR derived phenotype

项目摘要 为了响应(PAR-18-896)，这项提议的总体目标是充分发展 统计学家、医学信息学家、临床医生共同努力，专注于开发 通过现代知识工程和数据驱动实现严格的偏差纠正框架 统计建模，用于提高卫生系统数据的无偏性和重复性 推动研究。 在本方案中，我们将集中于：(1)开发一种新颖的先验知识引导的综合 似然方法，通过结合先前的表型准确度来实现偏差校正。(2) 开发方法和算法来解释结果和结果中的EHR表型错误 预测者。(3)验证、应用和软件开发。我们将使用建议的 对几个EHR数据集的偏差校正方法，以复制现有发现并进行调查 德克萨斯大学和宾夕法尼亚大学的多个数据集中的新假设。我们 还将为建议的方法开发软件，以促进正在进行的基于电子病历的临床 学习。

Project Summary In response to the (PAR-18-896), the overarching goal of this proposal is to fully develop a joint effort between statisticians, medical informaticians, clinicians with a focus on developing a rigorous bias correction framework through modern knowledge engineering and data-driven statistical modeling, for improving the unbiasedness and reproducibility of health system data driven research. In this proposal, we will focus on: (1) Develop a novel prior-knowledge-guided integrated likelihood approach to enable bias correction by incorporating prior phenotyping accuracy. (2) Develop methods and algorithms to account for EHR phenotyping errors in both outcomes and predictors. And (3) Validation, Application and Software development. We will use the proposed bias correction methods to several EHR datasets to replicate existing findings and investigate new hypothesis in multiple datasets at University of Texas and University of Pennsylvania. We will also develop software for the proposed methods to facilitate ongoing EHR-based clinical studies.