EAGER: Mining Heterogeneous Network Constructed from Multiple Data Sources

由于缺少涉及其他潜在实体及其关系的模式，依赖单一数据源进行知识发现通常会导致性能不理想。这在药物警戒等医疗保健信息学领域尤为重要。由于药物不良反应的影响，药物警戒是一个重要的卫生保健问题。它使患者的医疗状况复杂化，增加住院率，并导致更多的发病率和事件死亡。在目前的药物警戒研究中，大多数工作仅考虑单一数据源来发现两个实体之间的关联，即药物和药物不良反应。该项目开发了一个新的框架来集成多个数据源，包括自发报告系统、电子健康记录、药物数据库、科学文献和网络数据，用于异构网络挖掘。这种异构网络由多个实体组成，包括药物、药物不良反应、患者、疾病和症状，以及这些实体之间的各种类型的关系。该项目通过整合用于药物警戒应用的多个数据源，扩展了机器学习、数据分析和药物警戒的能力。特别是，在网络上纳入以患者为中心的数据创造了可能无法从主要由卫生专业人员提供的传统数据源获得的见解。该项目的成果包括异构路径挖掘和结构拓扑模式挖掘技术，用于四种药物警戒应用，即药物不良反应检测、药物-药物相互作用、处方级联和表型信息发现。这些技术也可以扩展到药物重新定位和标签外使用识别。本研究结果对药剂学、医学、公共卫生和计算机等多个学科有益。综合教育计划包括将研究成果纳入健康信息学理学硕士课程。外展计划包括组织讲习班、会议和研讨会，以传播研究成果。

Relying on a single data source for knowledge discovery often results in unsatisfactory performance because of the missing patterns involving other potential entities and their relationships. This is particularly important in healthcare informatics areas such as pharmacovigilance. Pharmacovigilance is an important healthcare issue due to the impact of the adverse drug reactions. It complicates patients' medical conditions, increase hospital admissions, and contribute to more morbidity and event death. In current pharmacovigilance research, most work only consider a single data source for discovering the associations between the two entities, namely drugs and adverse drug reactions. This project develops a novel framework to integrate multiple data sources, including spontaneous report systems, electronic health records, pharmaceutical databases, scientific literature, and web data, for heterogeneous network mining. Such a heterogeneous network consists of multiple entities, including drugs, adverse drug reactions, patients, diseases, and symptoms, and various types of relationships among such entities.This project extends the capability of machine learning, data analytics, and pharmacovigilance by integrating multiple data sources for pharmacovigilance applications. In particular, the inclusion of patient-centric data on the web creates insights that may not be obtained from traditional data sources mainly contributed by health professionals. The outcomes of the project include techniques for heterogeneous path mining and structural topological pattern mining on four pharmacovigilance applications, namely adverse drug reaction detection, drug-drug interaction, prescribing cascade, and phenotypic information discovery. Such techniques can also be extended for drug repositioning and off-label use identification. The result of this research is beneficial to multiple disciplines including pharmacy, medicine, public health, and computing. The integrated education plan includes incorporating the research findings into courses offered by the Master of Science program in Health Informatics. The outreach plan involves organizing workshops, conferences, and seminars to disseminate the research outcomes.