耐多药肺结核治疗过程中药物不良反应发生率高，危害严重，主动监测十分必要。采用传统流行病学调查耗时费力，对药品风险管理具有滞后性。电子病历中的海量数据为快速主动监测提供了可能，但如何从中提取有效的最小数据集是制约其使用的瓶颈。本研究旨在建立耐多药肺结核的通用数据模型，并评价其在不良反应信号发现中的真实性和收益。拟利用本课题组已有的全球基金耐多药肺结核控制项目多家医院的千余份电子病历，通过对比结构、标化术语，参考国际新近提出的模式，初步构建包括药物暴露、可疑不良事件、混杂因素等变量的通用数据模型；并根据该模型提取病历中的上述信息，采用比值失衡等方法评价其在不良反应信号挖掘中的真实性和可靠性；用高维倾向评分法控制混杂，筛选变量，优化模型；最后通过盖茨基金项目的电子病历评价该模型的收益。此外，本研究还将探讨不良反应信号检测、混杂控制等方法在电子病历数据挖掘中的应用，为主动监测提供新的思路和方法。

The active surveillance of Multidrug-resistant tuberculosis (MDR-TB) treatment is necessary because of its high incidence of adverse drug reactions (ADRs) in treatment and serious harm. The traditional epidemiological investigation is time-consuming, with a lag of drug risk management. Huge amounts of Electronic Medical Records (EMR) data provides possibility for quick active surveillance, however, how to extract effective and minimum data restricts its application. This study aims to establish a common data model (CDM) of MDR-TB, and evaluate its validity and benefit of ADRs signal detection. The research is intended to preliminary build CDM includes drug exposure of interest, possible adverse events, confounding factors variables by comparing the structure, standardizing the terminology and consulting latest proposed model in the world base on the existing EMR datasets, which are more than 1000, from Global Fund of MDR-TB Control Project hospitals; Extract the information of EMR according to the CDM, and then evaluate its validity and reliability by Proportional Reporting Ratio Method; Control confounding by High-Dimensional Propensity Score Method and then screen variables to optimize CDM; Finally evaluate the benefits of the MDR-TB CDM through the EMR datasets of Gates Foundation Project. In addition, this study also explores the application of ADRs signal detection methods and confounding control methods in EMR data mining, and provide new thinking and method of active surveillance.