Research On Mining Adverse Drug Reaction Signals Based On Association Rules Analysis

Adverse Drug Reaction(ADR)is an increasingly serious public health safety issue which endangers people’s property and lives.The Spontaneous Reporting System(SRS)is currently the main data resource for implementing post-marketing drug pharmacovigilance in various countries.ADR signal detection based on SRS report is divided into two aspects: monotherapy and combined medication: the methods of monotherapy signal mining include frequency method(PRR,ROR,MHRA,etc.)and Bayesian method(IC,etc.),the methods of combined medication signal mining include baseline models(additive model and multiplicative model).These classic signal detection methods have been adopted and successfully applied in various countries,but they also have the following disadvantages: high requirements for data quality,poor consistency of results,weak comprehensive performance,and lack of application of data mining technology.Hence this paper applies association rule analysis technology to our country’s spontaneous report data set to improve the comprehensive performance of signal detection and to provide technical support for our country’s pharmacovigilance.Firstly,this paper takes the SRS report from 2011 to 2018 provided by Adverse Drug Reaction Monitoring Center of Jiangsu Province as data resource.On the basis of using the World Health Organization Adverse Reaction Terminology(WHOART)to standardize the terms of the data,we preprocess data by extracting,splitting,and deduplicating them.We establish a database of adverse drug reactions about monotherapy and combined medication.Meanwhile,a known signal library is established as an objective criterion for the performance evaluation of the method proposed in this paper that is based on manual data retrieved from the Internet resources Yaozhi.com and Drugs.com and the DDI(Drug-Drug Interaction)adverse reaction data of specific drug combination.Secondly,this paper proposes a method named SDCSAR(Signal Detection of Clustering Stratification and Association Rules)which is based on clustering stratification and association rules in the field of monotherapy.The main steps of method SDCSAR include: clustering drugs in the data set according to the frequency of adverse reactions caused by different drugs;determining the minimum support and minimum confidence thresholds by MHRA and KS value(8)(6(True Positive Rate-False Positive Rate));using the determined thresholds for association rules to detect the ADR signals of each cluster and the overall data set.Finally,a new association rule-based model BAC(drug B→ adverse reaction A∩ drug C)is proposed in the field of combination medication,which uses promotion and certainty as rule generation indicators.It is compared with the two models in the baseline model(additive model,Multiplication model)and with the traditional model ABC based on association rules(drug A∩ drug B→ adverse reaction C)to evaluate signal detection performance.This paper proposes a method SDCSAR to determine the minimum support and minimum confidence threshold of association rules.Compared with the non-hierarchical association rules,method SDCSAR can detect some rare ADR signals and get higher comprehensive evaluation index F value.The result shows that the threshold determination strategy and the stratification idea of method SDCSAR are effective.In the field of combination medication,this paper proposes a new signal detection model BAC based on association rules.The model BAC gets higher F value in comparison with the traditional modal ABC,addition model,and multiplication model,suggesting better signal detection ability.