2016-08-012024-05-18https://scholars.lib.ntu.edu.tw/handle/123456789/696365摘要：藥品引起的不良反應(drug-induced adverse events or adverse drug events (ADEs))在近十年的衛生政策管理中，引發了許多熱烈的討論，主要的原因在於藥品所引起的不良反應可能對病人健康造成的重大危 害及衍生的龐大醫療費用。目前被各國政府普遍採納的，完整的藥品上市後安全監視(post-marketing surveillance)的資料蒐集過程 中，至少要包含藥品安全訊號的發現(signal detection)、再確認(signal refinement)與評估(signal evaluation)等三個面向，在過去幾年内，包括美國與歐洲都重新思考採用大型電子化健康照護資料，如保險申報 資料檔與電子病歷系統，作為找出藥品安全訊號的資料來源，然而，這些方法都還在發展當中。我們將結合團隊内臨床藥學與資訊管理的專才，利用臺灣健保資料庫做為資料來源，發展一個新的資 料探勘技術(data-mining technique)來找出藥品所引起的不良反應，並改良目前方法的不足，包括時間 與事件發生區間的界定、病人合併症的考量、藥品暴露的進階測量(包括暴露的劑量與藥品合併使用（藥 品交互作用)），同時藉助國内一家大型醫學中心的電子病歷資料庫，進行資料探勘技術模型的驗證。我們計劃在前兩年選定美國Mini-sentinel與OMOP及歐洲EU-ADR所建議評估，具有重要藥品安全 監視與公共衛生意義的藥品不良反應，進行偵測方法的模型建立與驗證，並在第三年進行藥品不良反 應對於醫療系統資源消耗的計算。<br> Abstract: Drug-induced adverse events have been recognized as significant public health problems as they can cause iatrogenic harms to patients and significant healthcare burden to the society.To meet the need of post-marketing surveillance, several initiatives in the United States and Europe have started to explore new approaches (such as data mining) to facilitate earlier signal detection using electronic healthcare records as data source, including electronic medical records and claims database.However, there are several limitations in existing literatures. Firstly, the methodology for signal detection using population-based electronic healthcare records is evolving. Therefore, empirical studies on this field are very scarce. Secondly, there is lack of model validation using external database in most existing studies. Thirdly, to the best of our knowledge, there is a lack of population-estimation of drug-induced adverse events associated with drug-drug interactions. Fourthly, the measurement of drug exposure was mostly limited to binary variable. Fifthly, patients’ comorbidities were less taken into account when investigating drug-induced adverse events. Sixthly, very few studies have addressed the economic burden associated with drug-induced adverse events.To address these limitations, we plan to use the National Health Insurance Research Database (NHIRD) and electronic medical records of one medical center to conduct a series of studies. As it will be very challenging to manage a large amount of data like this, we plan to adopt the data-mining techniques used in the field of pharmacovigilance.This study aims to(1)develop data mining technique to identify drug-induced adverse events;(2)do model validation for the regression-based approach;(3)enhance the regression-based approach to study drug-induced adverse events resulted from drug-drug interaction and to control for unobservable confounding factors;(4)estimate healthcare burden associated with drug-induced adverse events.藥品上市後安全監視藥品引起的不良反應全民健康保險資料庫電子病歷資料庫資料 探勘疾病負擔醫療資源post-marketing surveillancedrug-induced adverse events (ADEs)National Health Insurance Research Database (NHIRD)electronic medical recordsdata miningdisease burdenhealthcare utilization