Adverse Drug Reaction Knowledge Integration And Application Research Based On Big Data Mining

Objectives:Adverse drug reaction (ADR) is one of the major problems in the field of medicine. ADRs significantly increase the length of stay, economic burden, and risk of death for hospitalized patients. ADRs even cause serious public health events, which seriously threaten the life and health of human beings. Therefore, scientific research communities and health administrative departments have paid much attention to how to identify and monitor ADRs effectively. Now, drug postmarketing surveillance is one of the most important ADRs detecting methods in the world.In China, how to improve the monitoring system of drug safety, and raise the level of drug safety monitoring have been one of the main tasks of the medical and health system reform.Traditional drug postmarketing surveillance is based on the data mining of spontaneous adverse event reporting system. Nowadays, in the background of medical big data, how to integrate adverse drug event reporting system with different ADRs-related resources, then accomplish ADRs identification, storage, ADRs knowledge organization and utilization have not only become a hot topic in the academic community, but also become a key issue in public health administration departments all over the world. In this paper, we discuss the ADRs information management process base on the basic theory of Medical Informatics. We try to describe the ADRs knowledge discovery, integration and utilization process by the big data research technology and method. The final research objective is to provide knowledge for the medical decision support.Methods:First, we propose an ADRs big data information management theoretic model base on the DIKW theory by the literature review. And we regard this model as the theoretical framework of the whole paper. Then we conduct the data normalization for FDA Adverse Event Reporting System (FAERS) by natural language processing (NLP). The normalized dataset, AERS-DM, is stored by Elasticsearch, a distributed search engine for big data storage and retrieve. In next step, we detect the ADR signals for the most used 20 chronic disease treatments in the US pharmaceutical market in 2013. And we also design a data mining algorithm to detect those ADR signals which associate with sex difference. All the signals are stored in a ADR signal dataset. Visualization diagrams are also used for showing all the data mining results. Then we use NLP to integrate AERS-DM and ADR signal dataset with other ADRs-related resource such as RxNorm, NDF-RT, MedDRA, DrugBank and DART and to establish an ADRs knowledge base. Finally, we build an ADRs knowledge base retrieval system which base on Elasticsearch and CGI script programming to satisfy different user’s requirements. Achieving the goal of ADRs knowledge sharing and utilizing.Results:We build an ADRs big data information management theoretic model based on DIKW theory. The model summarized the law of information flow in the study of ADRs information management. Through the data mining, we detect 20237 ADR signals and 736 ADR signals with sex difference for 643 chronic disease drugs. All the signals are organized in a dataset. We also build an ADRs knowledge base includes AERS-DM, ADR signal dataset, RxNorm, NDF-RT, MedDRA, DrugBank and DART by knowledge integration. Finally, we establish ADR knowledge database retrieval system, which includes basic data retrieval module, clinical signal retrieval module and protein knowledge retrieval module 3 main functional modules and 9 sub functional modules, to satisfy the different users" knowledge demands on ADRs.Conclusions:(1) According to the ADRs big data information management theoretic model, we think the process of ADR data information management is actually the transformation of ADR data-information-knowledge-wisdom. The transformation process is achieved through different information activities. Its ultimate goal is to provide decision support for patients, doctors, researchers, pharmaceutical companies and drug administration departments, and finally to generate the ADRs-related wisdom. The level of data structure, information analysis and information value of ADRs-related data are upgraded during the process of ADRs big data information management.(2) We improve the quality of drug adverse event data through the data cleaning and normalization of FAERS. Elasticsearch search engine significantly improves the efficiency of ADRs big data storage and retrieval. Therefore, under the support of normalized dataset and Elasticsearch, the quality and efficiency of the ADRs data mining analysis are guaranteed.(3) We detect a lot of potential associations between drugs and adverse events through the big data mining of ADR signals. It provides information support for the research of clinical trials, pharmacological tests and epidemiological surveys. The ADRs-related data transform into information after big data mining.(4) We build a comprehensive ADRs knowledge base after the integration of different ADRs-related resources. Different ADRs-related resources are connected after integration, which expand the scope and content of ADRs-related information. Finally, we achieve the transformation of ADRs information to knowledge.(5) The ADRs knowledge base retrieval system can meet the different users" demands on ADRs knowledge. The retrieval system achieves the goal of ADRs knowledge sharing and utilizing. And accomplished the task of decision support for patients, doctors, researchers, pharmaceutical companies and drug administration departments. Finally, it laid the foundation for the ADRs wisdom.