Multi-source Data Based Data Mining Of Adverse Drug Reactions

Objectives: In this study,data mining on literature data,domestic adverse drug reaction reports and AERS-DM was conducted by using network analysis,statistical analysis and data mining algorithms to analyze the characteristics of adverse reactions of hypolipidemic drugs in China and abroad and find potential adverse reactions of hypolipidemic drugs.The objective of this study is to implement the integration of data mining results based on adverse drug reaction reports in China and abroad and Chinese literature data by using the drug adverse reaction ontology.So the potential adverse hypolipidemic drug reactions that have been reported abroad and not been found in China could be found out.Methods:(1)All the literatures in Wanfang Med Online and the literatures of the medical and health journals in the VIP database in 2012 were collected as the data source.After extracting keywords in the literatures by Perl,pairing keywords in the same document to construct a keyword network and filtering the cardiovascular drug sub-network,then the OCVDAE was integrated to find potential adverse drug reactions.(2)Data cleaning was conducted including extracting the adverse reaction terms,removing the stop words and deduplicating after downloading the data of hypolipidemic drug use and adverse drug reaction from the NCMI website.Next the data was normalized based on OAE,so that adverse reactions could be presented in standardized terms.Finally,statistical analysis was performed on the demographic characteristics,occurrence of most adverse reactions and serious reactions.(3)Extracting hypolipidemic drug data encoded in Rx Norm and adverse reaction data encoded in Med DRA from AERS-DM.Analyzing the occurrence of adverse reactions to hypolipidemic drugs abroad,and then calculating the PRR value of hypolipidemic adverse drug reactions to obtain signals of adverse hypolipidemic drug reactions.(4)The drug information was mapped to the Rx Norm code in the data from the literature database and NCMI and OAE terms were mapped to the Med DRA code to achieve the integration of the mining results.Results: Keywords network contains 380,684 nodes and 3,446,697 pairs of keywords.The cardiovascular drug sub-network included 111 cardiovascular drugs,7,223 nodes and 13,148 keyword edges.The maximum co-occurrence frequency of keyword pairs is 81.The feasibility of this method was verified by the case of telmisartan and association between dopamine and occurrence of phlebitis was found.In NCMI 843 reports were received,including 16 hypolipidemic drugs and 76 related reports.After adverse reactions were normalized and classified using OAE,a total of 4,053 adverse reactions records were obtained.In AERS-DM,there are 70,630 reports,331,543 records and 7,143 adverse reactions in 26 categories associated with 16 hypolipidemic drugs.Through the calculation of the PRR value,2008 adverse reaction signals of hypolipidemic drugs were obtained.Statistical analysis showed that the proportion of men was slightly larger than that of women and the most adverse reactions occurred in the 50-60 age group.The most reported route of administration in China is intravenous drip.Symptoms of anaphylactic shock were the most common symptoms in China and AERS-DM signals show that the most common adverse reaction was muscle-related diseases.After the mining results were integrated,a total of 1,929 signals were obtained.Then the known adverse reactions in OCVDAE were taken out,a total of 1877 potential adverse hypolipidemic adverse reaction signals were obtained.The results of knowledge discovery were verified in literatures.Conclusion: On the basis of multi-source data,this study conducted data mining research on the adverse reactions of hypolipidemic drugs at home and abroad.The feasibility of knowledge aggregation and knowledge discovery methods based on keyword networks was confirmed and keyword networks were displayed through visual graphs.At the same time,the credibility of signal monitoring of adverse drug reactions based on the PRR algorithm was confirmed.A method for data integration of adverse drug reactions based on ontology was proposed.The similarities and differences of cardiovascular adverse reactions in China and abroad were discovered,which provides guidance for the safe use of drugs and prevention of adverse reactions.