TRiPOD: Toward Reusable Phenotypes in Observational Data for AD/ADRD - managing definitions and correcting bias

项目摘要 电子健康记录（EHR）和医疗索赔等大型观察数据 成为促进包括阿尔茨海默氏症在内的临床和转化研究的有利来源 阿尔茨海默病相关性痴呆（AD/ADRD）。一个主要挑战是 进行观察性AD/ADRD研究是关于表型的-缺乏集中的 用于托管和标准化AD/ADRD研究中表型定义的存储库，以及一些 已经开发了一些方法来解决与观察中的表型错误相关的偏倚 数据因此，本提案的总体目标是充分开展以下方面的共同努力： 医学信息学家，统计学家，临床医生和流行病学家，重点是建立一个 一套严格的方法和工具，用于管理表型定义和纠正 通过现代知识工程和数据驱动的统计， 建模为了实现这一目标，我们在本研究中提出了三个具体目标：（1）目标1 -收集， 标准化，并共享AD/ADRD观察性研究中使用的常见表型的定义 研究;（2）目标2 -开发新的算法，以纠正与表型错误相关的偏倚 当用户将现有的表型定义应用于本地数据时;以及（3）目标3 -重新定义，细化， 通过示范研究和社区参与传播拟议的方法和工具。 我们相信，这里提出的信息学方法和工具将改善目前的做法， 表型数据的管理和分析，从而提高再现性和质量 AD/ADRD的观察性研究。

Project Summary Large observational data such as electronic health records (EHRs) and medical claims have become an enabling source for facilitating clinical and translational research including Alzheimer's Disease and Alzheimer's Disease Related Dementia (AD/ADRD). One major challenge for conducting observational AD/ADRD studies is about phenotyping – there is a lack of a centralized repository for hosting and standardizing phenotype definitions in AD/ADRD research and few methods have been developed to address bias associated with phenotyping errors in observation data. Therefore, the overarching goal of this proposal is to fully develop a joint effort between medical informaticians, statisticians, clinicians, and epidemiologists with a focus on building a rigorous set of methods and tools for managing phenotype definitions and for correcting bias in observational data analysis, through modern knowledge engineering and data-driven statistical modeling. To achieve that goal, we propose three specific aims in this study: (1) Aim 1 - Collect, normalize, and share definitions of common phenotypes used in AD/ADRD observational research; (2) Aim 2 - Develop novel algorithms to correct bias associated with phenotyping errors when users apply existing phenotype definitions to local data; and (3) Aim 3 - Validate, refine, and disseminate proposed methods and tools by demonstration studies and community engagement. We believe informatics methods and tools proposed here will improve current practice on phenotypic data management and analysis, thus enhancing the reproducibility and quality of observational studies on AD/ADRD.