| ABSTRACTWith the development and improvement of electronic medical records, it become possible to carry out active drug safety surveillance based on large scale electronic healthcare database. We summarize the data mining methods of detecting adverse drug reaction signals based on electronic medical database, including disproportionality analysis, traditional pharmacoepidemiological designs, prescription sequence symmetry analysis (PSSA), sequential statistical testing, temporal association rules, supervised machine learning (SML), and the tree based scan statistic. And we also review the methodology researches of evaluating these methods. When considering the performance of these methods, the self controlled designs, the PSSA, and the SML seemed the better approaches. When considering using the methods of signal detection for the routine drug safety active surveillance, whether the results will be interpreted confidently and the method principle is easy to understand, as well as whether the method can provide the signal strength or if the method is easy to implement are the key factors that affect its practical application. |
