Discovering and Applying Knowledge in Clinical Databases

描述（由申请人提供）：我们正在进行的项目“在临床数据库中发现和应用知识”的长期目标是从电子健康记录（EHR）中的数据中学习，并将这些知识应用于相关问题。电子健康记录（EHR）的出现大大增强了开展观察性研究的能力，在长期纵向研究中开辟了覆盖新出现的问题、不同人群、罕见疾病和慢性病的可能性。不幸的是，电子病历带来了额外的挑战。我们认为，最大的挑战来自于医疗保健过程记录的不准确性、不完整性、复杂性和由此产生的偏见。换句话说，电子病历数据不是简单的研究数据，有更多的噪音和缺失的一些值；相反，电子病历存在系统性偏见，必须在数据发挥其潜力之前加以解决。我们建议描述医疗保健过程对电子病历数据的影响，列举潜在的偏差，并提供规避它们的机制。实际上，我们建议将电子病历本身作为一个感兴趣的对象来研究，使用新的模型、数据挖掘、现有的知识库和创新的算法来更好地理解电子病历偏差，以便我们能够识别它们，纠正它们或避免它们。我们包括来自两个国家主要表型项目的专家，eMERGE和OMOP。

DESCRIPTION (provided by applicant): The long term goal of our ongoing project, "Discovering and applying knowledge in clinical databases," is to learn from data in the electronic health record (EHR) and to apply that knowledge to relevant problems. The advent of the electronic health record (EHR) greatly amplifies the ability to carry out observational research, opening the possibility of covering emerging problems, diverse populations, rare diseases, and chronic diseases in long-term longitudinal studies. Unfortunately, the EHR carries additional challenges. We believe that the biggest challenge comes from the inaccuracy, incompleteness, complexity, and resulting bias inherent in the recording of the health care process. Put another way, EHR data are not simply research data with more noise and missing some values; instead the EHR carries systematic biases that must be addressed before the data can reach their potential. We propose to characterize the effects of the health care process on EHR data, to enumerate the potential biases, and to provide mechanisms to circumvent them. In effect, we propose to study the EHR as an object of interest in itself, using new models, data mining, existing knowledge bases, and innovative algorithms to better understand EHR biases so that we can identify them and correct them or avoid them. We include expertise from two of the nation's major phenotyping projects, eMERGE and OMOP. We hypothesize that we can learn about biases due to the health process through data mining and knowledge engineering and that we can correct or at least avoid those biases, enabling us to better answer informatics and clinical questions. Our aims are as follows: (1) Study health care process biases by correlating raw EHR variables with a panel of health care process-related variables (e.g., admission), using lagged correlation to account for temporal effects, and populating a health care process resource with the correlations and observations. (2) Find associations among raw EHR variables using lagged correlation, information theory, Granger causality, and temporally ordered N-tuples of events, correcting for the health care process biases discovered in Aim 1. (3) Facilitate the definition of higher-level clinical phenotype concepts by applying knowledge resources-including eMERGE and OMOP phenotype definitions and ontologies such as our Medical Entities Dictionary and the UMLS-to the fruit of Aims 1 and 2 to produce semi-automated and automated phenotype query definitions. (4) Develop a high-throughput method to validate phenotype definitions by measuring the ability to uncover known associations, use the generated phenotypes and associations to answer clinical questions, and disseminate the results, including a large knowledge base of correlations that can be used by other researchers to conduct their own studies.