2017-08-012024-05-13https://scholars.lib.ntu.edu.tw/handle/123456789/651266摘要：背景隨著巨量資料時代的來臨，利用電子病歷來進行文字探勘是近年來熱門的研究主題。電子病歷包含豐富的資訊，利如症狀、癥侯、臨床醫師的判斷、檢驗結果、與治療的反應；文字探勘技術即是結合資料探勘、自然語言處理、與機器學習等技術的方式，讓電腦自動化的分析電子病歷內容中未結構化的出院病摘部分。同時利用文字探勘的方式，亦可去除可辦識的個人資料，有助於保護個案的隱私。方法1. 資料來源為曾於台大醫院精神醫學部病房住院，並有精神疾病診斷之出院病歷2. 由兩位精神科專科醫師，回顧電子病歷，並確定其症狀、診斷、生活習慣、用藥物資料。3. 先利用半數的電子病歷進行文字探勘技術的訓練。再利用半數的病歷進行驗正；並計算其準確性與敏感度4. 應用電子病歷文字探勘的資訊。(1)進行檢驗健保申報疾病碼的準確性；(2)運用入院病歷資料，預測其治療效果。預期結果與影響性運用電子病歷文字探勘的技術，可大符減少人工病歷回顧的時間，並可進行大規模的資料收集、分析與驗證。以本計畫所提出的兩項應用來說，可驗證目前健保申報疾病碼的準確性；並用於預測治療成效。對於學術發展而言，未來如果能夠結合電子病歷的文字探勘技術與其他醫療巨量資料(如健保資料庫)，可突破目前次級資料中，缺乏症狀的記錄、疾病嚴重度、整體功能、生活型態、心理社會壓力等研究限制。對整體公共衛生而言，更能夠提供影影響治療效果之參考資料，以利於疾疾的治療與預防。<br> Abstract: BackgroundIn the era of big data, the number of electronic health records (EHRs) increased dramatically in therecent years. Information in EHRs includes patient-report symptoms and sign, clinical judgment ofphysicians, laboratory and image results, and treatment response and complications. Manual chart reviewcould obtain comprehensive information from EHRs; however, it is quite time-consuming and labor-intensive.Text mining approach can be applied to automatically retrieve and extract information from the unstructuredpart of EHRs. Text mining is a combination of the development of information retrieval, natural languageprocessing, information extraction and data mining. There are several advantages of text mining in EHRs.First, automatic information extraction becomes very economic; therefore, it is possible to deal with largevolume of information. Second, using de-identification techniques could protect patient’s privacy. Allpersonal information could be deleted and keep useful clinical information only. Third, use ofmachine-learning techniques might further facilitate active learning for identifying important clinical featureswith minimal cost and labors.Method1. We will conduct a chart review and randomly select 600 discharge note from the National TaiwanUniversity Hospital (NTUH) EHR system from January 1, 2006 to September 30, 2016.2. Two board-certified psychiatrist will review these EHRs as gold standards. All personal information willbe de-identified during text mining processing.3. The training set consists of 300 discharge notes and another independent 300 discharge notes will beselected as a test dataset to assess the performance of text mining methods.4. We will apply the text mining technique to validate of the diagnosis of psychiatric disorder in claimsrecords and to explore predictors of treatment response and prognosis among those with psychiatrichospitalizationAnticipated results and significanceUsing text mining could markedly reduce the time and labor for manual chart review. Combination textmining techniques and other health-related big data, such as claims database of National Health Insurancecould overcome the limitation of study using secondary database, including lack of information for symptomprofile, severity, function, lifestyle behaviors, and psychosocial stressors, etc. For the public health, the use oftext mining in electronic health information could provide knowledge for improving disease treatment,prediction of clinical outcome, and prevention.