Research On Drug Risk Assessment Based On Clustering Analysis

Adverse Drug Reaction(ADR)is a harmful and unexpected reaction,which harms human health and even endangers life to varying degrees.Therefore,the ADR monitoring of post-marketing drugs has received increasing attention from countries around the world and has become the top priority of pharmacovigilance.Existing researches were mainly based on the monitoring data of adverse drug reactions,using various signal detection methods to mine the correlation between drugs and adverse reactions and timely risk warning.They mainly focused on individual drug and single adverse reaction,lacking of the signal mining of drug,especially the drugs risk classification from the perspective of ADRs.To this end,this paper proposes a drug risk assessment method based on cluster analysis to mine the key features of drug risk level classification and drug classes,in order to provide decision-making reference for China's pharmacovigilance work.This paper selects a total of 400 adverse reactions caused by 76 antibiotics in the ADR surveillance data of the China Food and Drug Administration(CFDA)from 2010 to 2011 as research data.These data are combined with WHO-ART adverse reaction terminology and drug specifications to standardize the original data,and extract record attribute values such as report type(general,severe),drug category,drug name,and adverse reaction name as the experimental data set.(1)Research on signal detection methods: The research study the mainstream signal detection methods and apply them to ADR monitoring data of China.After comparative analysis,PRR,IC,and binary value are selected as the signal detection methods for this study and a vector space model(VSM)is established using the pre-processed data set in three methods.Then the research use the elbow rule to determine the optimal number of clusters for various signal detection methods.(2)Application research of clustering algorithm: According to the characteristics of the selected ADR monitoring data,the research takes drug names as the object and features adverse drug reactions,and uses the Fuzzy C-Means(FCM)clustering algorithm to experimentally simulate antibiotic drugs based on three signal detection methods and classifies the drugs into three categories.Then the research designs and uses the classification accuracy evaluation function to determine the PRR method as the best signal detection method.(3)Risk grade evaluation study: The research proposed an injury index based on the severity of adverse reactions.Based on the calculation of the injury index of the three types of antibiotic drugs,the corresponding risk level was obtained.The study verified the credibility of the results with information such as the number of prescription and over-the-counter drugs in each risk level and the proportion of serious adverse reactions.(4)Risk feature extraction research: The research uses TF-IDF(Term Frequency–Inverse Document Frequency)algorithm to extract the top 10 key adverse reaction characteristics of each class and based on the WHO-ART correlation involving system-organs for risk assessment of each class.It further verifies the correctness of the correspondence between risk characteristics and risk levels.This paper takes antibiotic drugs of the ADR monitoring data in China as the research object,and uses cluster analysis and feature extraction technology to achieve ADR-based drug risk classification and risk feature identification.It provides a reference method for establishing drug risk classification mechanism and a new perspective for the reuse of ADR monitoring report data resources in China.