An Information Fusion Approach to Longitudinal Health Records

描述（由申请人提供）：我们的目标是利用信息融合方法来整合结构化和非结构化信息，以生成纵向健康记录（LHR），以加快招募患者进入临床试验的步伐。由于电子健康记录（EHR）包含患者临床病史的临床摘要，人们会认为可以很容易地利用它们来自动筛选和识别潜在的合格患者。然而，大多数电子病历没有很好地设计来支持合格患者的筛选，并且由多个数据源组成，这些数据源通常是冗余或不一致的，存储在不协调的非结构化临床叙述和结构化数据中。这些特点使得电子病历难以用于将患者与临床试验方案的复杂事件和时间标准相匹配。本研究提出一个改进的LHR，它包含了一个病人的全面临床总结，可以提高病人的筛查。我们建议使用一种信息融合的方法来生成这种LHR，它将来自多个数据源的信息合并在一起，解决了数据的含义和时间性质，这样得到的信息比单独使用这些数据源时可能得到的信息更准确。

DESCRIPTION (provided by applicant): Our goal is to leverage an information fusion approach to integrate structured and unstructured information to generate a longitudinal health record (LHR) for accelerating the pace at which patients can be recruited into clinical trials. Because electronic health records (EHR) contain clinical summaries of a patient's clinical history, one would assume that they could be easily leveraged to automatically screen and identify potentially eligible patients. However most EHRs are not well designed to support screening of eligible patients and are composed of multiple data sources that are often redundant or inconsistent, stored in uncoordinated unstructured clinical narratives and structured data. These characteristics make EHRs difficult to use for matching patients against the complex event and temporal criteria of clinical trials protocols. This research proposes that an improved LHR, which contains a comprehensive clinical summary of a patient, can improve patient screening. We propose using a method of information fusion to generate this LHR, which merges information from multiple data sources, that addresses both the meaning and temporal nature of data, such that the resulting information is more accurate than would be possible if these sources were used individually. The specific aims are to: 1) characterize the barriers of using EHR sources for screening in terms of data redundancy, inconsistency, lack of structure, and temporal imprecision; 2) automatically extract information from unstructured EHR sources necessary for screening patients against clinical trials eligibility criteria using natural language processing; 3) developan LHR appropriate for screening patients against eligibility criteria using information fusion methods based on semantic and temporal information; and 4) evaluate the accuracy of an LHR formed through information fusion for screening patients against clinical trials eligibility critera. The respective hypotheses to be tested are: 1) Different parts of the EHR will contain variable amounts of redundancy, inconsistency, and temporal imprecision. Some sources will be more valuable for matching patients than others to clinical trials eligibility criteria. 2) Including th information contained in the unstructured notes will reduce the false positive rate of identifying potentially eligible patients over leveraging only the structured data in the EHR. 3) By using information fusion methods based on leveraging semantic and temporal information on a combination of structured and unstructured data, we will be able to accurately summarize the information contained in uncoordinated EHR data sources into an LHR that can be used for screening patients for clinical trials. 4) The use of information fusion to generate a longitudinal health record will increase the sensitivity and specificity of electronic clinical trial screening ver using a traditional EHR. With an LHR formed through information fusion for screening patients for clinical trials eligibilit, we will be able to not only reduce the amount of staff effort required to recruit a patient into a clinical trial, but also accelerate the pace at which clinical trials can be conducted.