Exploration Of Adverse Reaction Signal Detection Method Based On FAERS Database

In the past decades,adverse drug reactions have caused significant human and financial losses around the world,so pre-marketing clinical trials of target drugs are indispensable.However,clinical trials can only start from the chemical composition of drugs,and the research time is limited,the sample is few and other factors.Therefore,it is necessary to set up a Spontaneous Report System(SRS)of the drug and its adverse reactions after the drug has passed the clinical trials,as well as carry out statistical analysis simultaneously.This study selected data from the FDA Adverse Events Reporting System(FAERS)database from 2015 to 2022,and preprocessed and statistically analyzed the age data.In addition,based on the 399 pairs of drug and adverse reaction gold standards provided by the Observing Medical Outcomes Partnership(OMOP),each drug and reaction was analyzed in the pre processed FAERS data to obtain corresponding statistical data.The classical methods of drug and adverse reaction analysis are mainly based on frequency method and Bayesian method.This paper comprehensively analyzed the signals associated with399 pairs of drugs and adverse reactions based on FAERS and OMOP data.According to Receiver Operating Characteristic(ROC)curve,It can be seen that the AUC(Area Under Curve)and Youden Index of the two Bayes-based methods are higher than those of the two methods of frequency method in acute kidney injury and acute liver injury responses with large statistics.However,there was no significant difference between the two methods in acute myocardial infarction and chronic gastrointestinal bleeding.In order to fully explore the correlation between drugs and adverse reactions,this paper proposes two deep forest-based association analysis methods for drugs and adverse reactions,namely,deep forest-based integrated feature method and deep forest-based fully correlated feature screening method.This kind of deep forest-based approach is a deep structure composed of multiple random forests.The results showed that the AUC value and Youden index of the deep forest-based integrated feature method were higher than those of random forest in four kinds of adverse reactions,especially in acute kidney injury response,the AUC value and Youden index were higher than that of random forest by 0.03 and 0.02.The full-correlation feature screening method based on deep forest has a good performance in the response with fewer statistics.Compared with the deep forest integrated feature method,the AUC value of acute myocardial infarction is increased by 0.06,and the Youden index is increased by 0.06.In addition,both methods based on deep forest are superior to the classical methods of drug and adverse reaction analysis.In this paper,an ADR reporting and inquiry system is developed.Ordinary users and medical management personnel can conduct data query through the algorithm model embedded in the background of the system to understand the correlation between drugs and adverse reactions.In addition,management personnel can conduct unified management of drugs and adverse reactions through the background,which is convenient to upgrade the database and algorithm later.