2014-08-012024-05-14https://scholars.lib.ntu.edu.tw/handle/123456789/659759摘要：研究背景：大型健康照護資料庫累積了大量病人的診療紀錄，遠超過任何一位醫師或任何一家醫院的經驗。如何善用這些資訊增進健康，協助個別病人醫療決策，改善醫療照護成果，是目前最重要的課題。 研究目的：以資料探勘技術與貝氏模型分析巨量健康資料，建立第二型糖尿病患在接受 metformin 第一線藥物治療後，接受不同第二線口服降血糖藥物 (sulfonylurea、glinide、alpha-glucosidase inhibitor、pioglitazone、DPP4-inhibitors) 一年內發生嚴重心血管事件，包括心肌梗塞、心衰竭、腦中風之機率預測模型。 研究方法：本研究為回溯性追蹤研究。將以 2000/1/1-2010/12/31 年齡 20 歲以上之第二型糖尿病，第一線 metformin 治療後接受第二線口服降血糖藥物患者為研究對象，包含的變項包括性別、年齡、罹患糖尿病時間、接受第一線 metformin治療時間、糖尿病併發症、共病、使用藥物、體重、血壓、及實驗室檢查結果。以資料探勘技術，根據上述資料的可能組合，針對病患發生心血管風險的機率進行適當分類，並估計每一組之發生率。進一步利用 Bayesian hierarchical rule modeling 建立初步預測模型，並以 2011/1/1-2012/12/31 病患資料驗證預測模型之準確度。 預期目標：應用資料探勘技術與適用於個人醫療的貝氏預測模型，建立以大型健康資料庫輔助個別病患醫療決策的方法。<br> Abstract: Background: Large healthcare databases compile tremendous amount of various patients' diagnostic and treatment records, far greater than the limited experiences of any physician or hospital. How to make the best use of these data to facilitate individual patient's therapeutic decision making and to improve outcome of care will become an important issue. Study goal: We will use data mining technique and Bayesian hierarchical rule modeling to analyze large healthcare database. A model will be built to predict 1 year incidence of severe cardiovascular events, including myocardial infarction, heart failure, and stroke, among individual type 2 diabetic patient receiving different oral anti-diabetic agents (sulfonylurea, glinide, alpha-glucosidase inhibitor, pioglitazone, DPP4-inhibitors) after metformin as the 1st-line treatment. Methods: A retrospective cohort study will be conducted by using the National Health Insurance claims database and clinical datasets in several hospitals in Taiwan. Participants will be type 2 diabetic patients, aged ≥ 20 years old, who initiated the above medications during 2000-2010 after 1st-line therapy of metformin. Potential explanatory variables will include age, sex, diabetic duration, metformin treatment duration, diabetes-associated complications, comorbidities, concomitant medications, body weight, blood pressure, and laboratory results. Pattern recognition to define subgroups with similar risks will be performed by using data mining techniques such as classification tree or support vector machine. Risk estimates will be used as the prior probabilities, and a prediction model for individual patient's risk while receiving different 2nd-line oral agents will be developed by using Bayesian hierarchical rule modeling. Data during 2011-2012 will be used for validating the accuracy of the proposed prediction model. Expected results: This study will apply data mining technique and Bayesian analysis to predict individual patient's outcome, and to build a framework of using large healthcare database to facilitate individual therapeutic decision making.Non-tuberculous mycobacteriuminterleukin-12interferon-γautoantibodyhost geneticsHuman leukocyte antigen (HLA)