Leveraging Electronic Health Records and Genomic Biobanks for Kidney Stone Disease

项目总结 肾结石是一种非常普遍和复发的疾病。我们目前对肾结石疾病危险因素的认识 疾病协会主要依赖于来自图表审查的数据、非特定的管理数据集和 观察研究的二次分析。目前的研究设计存在样本量小、异质性差的问题 患者群体，以及缺乏标准化的准确性数据和结果定义。被广泛采用的 电子健康记录为肾结石疾病提供了新的研究机会。EHR包含 一个强大的临床存储库，存储从临床护理中收集的数据。然而，目前有有限的 用于在电子病历中识别和表征肾结石患者的工具。这项研究的目标是建立 利用EHR数据研究肾结石的可行性。将EHR数据以高效和 可以设计具有成本效益的方式、自然语言处理和深度学习方法来识别 以及肾结石患者的表型和临床转归。我们未确认的EHR与一个DNA生物库有关联 这可以使研究基因与疾病的联系成为可能。这个项目有两个具体目标。在目标1中， 我们将在我们的EHR和连接的DNA生物库中进行遗传关联研究。我们将复制之前的 描述了与基因变异和肾结石疾病的联系。然后我们将在全基因组范围内进行 联想研究，以发现新的联想。在目标2中，我们的目标是开发和验证可计算的 从EHR中提取肾结石疾病临床结果的框架。临床上有意义的结果 包括症状性结石通过和放射学结石特征。我们将开发和测试Natural 用于提取临床关键字和上下文信息的语言处理和深度学习算法 笔记和报告。我们将使用手动注释作为黄金标准来训练和测试这些算法。这 AIM将使用结构化和非结构化EHR数据对每个肾结石患者进行严格的表型鉴定。 该项目的成功完成将为推进基因组医学和精确度奠定基础 健康支持肾结石患者的临床决策。

PROJECT SUMMARY Kidney stones are highly prevalent and recurrent. Our current understanding of kidney stone disease risk factors and disease associations has relied primarily on data from chart review, nonspecific administrative datasets, and secondary analyses of observation studies. Current study designs suffer from small sample sizes, heterogenous patient groups, and lack of standardized accuracy data and outcome definitions. The widespread adoption of electronic health records (EHRs) provides novel research opportunities for kidney stone disease. EHRs contain a robust clinical repository of data collected over time from clinical care. However, there are currently limited tools to identify and characterize kidney stone patients in the EHR. The objective of this study is to establish feasibility of utilizing EHR data to investigate kidney stone disease. To structure EHR data in an efficient and cost-effective manner, natural language processing and deep learning methods can be designed for identifying and phenotyping kidney stone patients and clinical outcomes. Our de-identified EHR is linked to a DNA biobank that can enable investigation of genetic associations with disease. This project has two specific aims. In Aim 1, we will perform genetic association studies in our EHR and linked DNA biobank. We will replicate previously described associations with genetic variants and kidney stone disease. We will then perform a genome-wide association study to discover novel associations. In Aim 2, our goal is to develop and validate a computable framework to extract clinical outcomes of kidney stone disease from the EHR. Clinically meaningful outcomes include symptomatic stone passage and radiographic stone characterization. We will develop and test natural language processing and deep learning algorithms to extract keywords and context-based information in clinical notes and reports. We will train and test these algorithms using manual annotation as the gold standard. This aim will enable rigorous phenotyping of each kidney stone patient using structured and unstructured EHR data. Successful completion of this project will lay the groundwork towards advancing genomic medicine and precision health to support clinical decision-making in kidney stone patients.