An in-silico method for epidemiological studies using Electronic Medical Records

描述：观察性流行病学研究是确定影响人群健康和疾病的因素以及确定癌症等疾病的最佳治疗方法的有效方法。然而，传统的流行病学研究通常涉及人员密集的工作（例如手工图表和公共记录审查），并且在获得结论性结果之前可能非常耗时。近年来，医院电子病历（Electronic Medical Records， EMR）系统积累了大量详细的纵向临床数据，成为流行病学研究的宝贵数据来源。然而，有两个障碍阻碍了电子病历数据在流行病学研究中的广泛使用。首先，电子病历中的大多数详细临床信息都嵌入在叙述文本中，人工提取这些信息的成本非常高。其次，电子病历通常存在数据质量问题，如选择偏差和缺失数据，这需要对随机对照试验开发的传统统计方法进行调整。

DESCRIPTION: Observational epidemiological studies are effective methods for identifying factors affecting the health and illness of populations, as well as for determining optimal treatments for diseases, such as cancers. However, conventional epidemiological research usually involves personnel-intensive effort (such as manual chart and public records review) and can be very time consuming before conclusive results are obtained. Recently, a large amount of detailed longitudinal clinical data has been accumulated at hospitals' Electronic Medical Records (EMR) systems and it has become a valuable data source for epidemiological studies. However, there are two obstacles that prevent the wide usage of EMR data in epidemiological studies. First, most of the detailed clinical information in EMRs is embedded in narrative text and it is very costly to extract that information manually. Second, EMRs usually have data quality problems such as selection bias and missing data, which require adaptation of conventional statistical methods developed for randomized controlled trials. In this study, we propose an in silico informatics-based approach for observational epidemiological studies using EMR data. We hypothesize that existing EMR data can be used for certain types of epidemiological studies in a very efficient manner with the help of informatics methods. The informatics-based approach will contain two major components. One is an NLP (Natural Language Processing) based information extraction system that can automatically extract detailed clinical information from EMR and another is a set of statistical and informatics methods that can be used to analyze EMR-derived data. If the feasibility of this approach is proven, it will change the standard paradigm of observational epidemiological research, because it has the capability to answer an epidemiological question in a very short time at a very low cost. The specific aim of this study is to develop an automated informatics approach to extract both fine-grained cancer findings and general clinical information from EMRs and use them to conduct cancer related epidemiological studies. We will perform both casecontrol and cohort studies related to prevention and treatment of breast and colon cancers using EMR data. The informatics approach will be validated on EMRs from two major hospitals to demonstrate its generalizability. Epidemiological findings from our study will be compared to reported findings for validation. Narrative: According to the American Cancer Society, about 7.6 million people died from various types of cancer in the world during 2007. It is very important to identify risk factors of cancers and to determine optimal treatments of cancers, and epidemiological study is one of the methods to achieve it. This proposed study will use natural language processing technologies to automatically extract fine-grained cancer information from existing patient electronic medical records and use it to conduct cancer related epidemiological studies, thus accelerating knowledge accumulation of cancer research. SPECIAL REVIEW NOTE: In order to conform to the scientific objectives outlined in the program announcement RFA-GM-09-008, EUREKA applications submitted to the NCI were initially evaluated by a group of reviewers representing diverse scientific interests. The priority score reflects the average of all the scores given by the full committee after a thorough discussion.