| Summary: | 碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 97 === Adverse Drug Reaction (ADR) is one of the most important issues on drug safety assessment and should not be ignored. In fact, many adverse drug reactions cannot be discovered through limited pre-marketing clinical trials. Instead, they can only be recognized by a long term of post-marketing surveillance of drug usages. In light of this, how to detect adverse drug reactions as early as possible is thus an important research topic in the pharmaceutical industry. Recently, the accumulation of large volumes of adverse events and the flourish of data mining technologies have encouraged the development of statistical and data mining methods for detecting ADRs. These stand-alone methods, without integration into knowledge discovery systems, are tedious and inconvenient to users and the processes of exploration are time-consuming. In this thesis, we thus propose an interactive system platform for ADRs detection. By integrating the concept of ADRs data warehouse and innovative frequent pattern mining techniques, the proposed system can not only support OLAP style of multidimensional analysis of ADRs, but also offer interactive discovery of associations between drugs and symptoms, called drug-ADR association rule, which can be further specialized by other factors interesting to users, such as demographic information. We consider four types of drug-ADR association rules, including associations of single drug and single symptom, multiple drugs (interactions) and single symptom, single drug and multiple symptoms, and multiple drugs and multiple symptoms. We propose two efficient algorithms to accomplish interactive discovery of the first two types of association patterns. Experiments indicate that interesting and valuable drug-ADR association rules can be efficiently mined. As to the third and the fourth types, we propose the concepts of mining methods; their works will complete in the future.
|
|---|
