Evidence-based Diagnostic Tools for Translational and Clinical Research (eTfor2)

描述（由申请人提供）：eTfor2项目将开发和评估开源程序和知识表示，以更好地表征转化和临床研究的患者。该项目涉及国家医学图书馆（NLM） RFA倡议：(a)信息和知识处理，包括自然语言处理和文本摘要，(b)连接表型和基因组信息的方法，以及(c)来自异构来源的信息集成。转化研究将临床患者描述符（表型组）与基因组调查结果联系起来，例如全基因组关联研究（GWAS）。定义表型的标准方法需要昂贵、劳动密集型的队列登记，以确定患有疾病的患者和适当的对照。最近，翻译和临床研究人员已经使用电子病历（EMR）数据作为识别患者特征的替代方法。然而，由于ICD9计费代码固有的不准确性，EMR案例提取需要大量的人工审查和案例选择的“调优”。虽然促进EMR文本提取的相关和有用的自然语言处理（NLP）方法已经激增，但这些方法采用的目标患者描述符通常仍然是非标准的和局部定义的，并且因疾病、项目和机构而异。在最好的情况下，这种NLP应用程序使用标准术语描述符（如SNOMED-CT）作为EMR提取目标。然而，目前还没有普遍使用的“标准”知识库，将这些“可提取”的描述符与学术质量的知识库联系起来，详细说明每种疾病中可靠报告的发现。为了促进转化和临床研究，eTfor2项目将提供一个开源的、基于证据的电子临床知识库（KB）和相关的NLP工具，使任何地点的研究人员能够在发现和疾病水平上提取一组基于emr的标准“目标”现象描述符。它将进一步包括诊断决策支持逻辑，以确认其电子病历中对患者诊断的支持程度。eTfor2项目将减少为临床和转化研究收集EMR患者描述符所需的工作量，并使新的转化工作能够在发现和疾病水平上确定基因组关联。eTfor2资源应提高基于电子病历的转化和临床研究的质量和跨机构有效性。

DESCRIPTION (provided by applicant): The eTfor2 project will develop and evaluate open-source programs and knowledge representations to better characterize patients for translational and clinical research studies. The project addresses National Library of Medicine (NLM) RFA initiatives for: (a) information & knowledge processing, including natural language processing and text summarization, (b) approaches for linking phenomic and genomic information, and (c) integration of information from heterogeneous sources. Translational studies correlate clinical patient descriptors (phenome) with results of genomic investigations, e.g., genome-wide association studies (GWAS). Standard methods for defining phenotypes require costly, labor-intensive cohort enrollments to identify patients with diseases and appropriate controls. Recently, translational and clinical researchers have used electronic medical record (EMR) data as an alternative to identifying patient characteristics. However, EMR case extraction requires substantial manual review and "tuning" for case selection, due to the inaccuracies inherent in ICD9 billing codes. While relevant and useful natural language processing (NLP) approaches to facilitate EMR text extraction have proliferated, the target patient descriptors these approaches employ typically remain non-standard and locally defined, and vary from disease to disease, project to project and institution to institution. At best, such NLP applications use standard terminology descriptors such as SNOMED-CT as EMR extraction targets. Yet, there is no generally utilized "standard" knowledge base that links such "extractable" descriptors to an academic-quality knowledge source detailing what findings have been reliably reported to occur in each disease. To facilitate translational and clinical research, the eTfor2 project will make available an open-source, evidence-based, electronic clinical knowledge base (KB) and related NLP tools enabling researchers at any site to extract a standard "target" set of EMR-based phenomic descriptors at both the finding and disease levels. It will further include diagnostic decision support logic to confirm the degree of support for patients' diagnoses in their EMR records. The eTfor2 project will decrease effort required to harvest EMR patient descriptors for clinical and translational studies, and enable new translational work that identifies genomic associations at both finding and disease levels. The eTfor2 resources should improve the quality and cross-institutional validity of EMR-based translational and clinical studies.