Research On ADR Outlier Detection Based On Context Condition

Drug safety is related to human health.The important content of pharmacovigilance in various countries is to monitor and manage the safety of post-marketing drugs.The mining of conventional signals is currently the main concern in the field of adverse drug reactions（ADR）.Because traditional signal detection methods depend on imbalance theory,there is a phenomenon of data obscuration,therefore,some outliers with important research value cannot be discovered by these methods.The aim of this paper is to explore potential and valuable outliers based on ADR monitoring data so as to help to make decision for pharmacovigilance in China.The main work includes the following aspects:1.Data processing:On the basis of cleaning up and standardizing the monitoring data set of ADRs of Jiangsu Province,taking the“drug-ADR”combination as the research object,the signal value of each combination is calculated as the eigenvalue of each object by using the international mainstream signal detection methods,such as PRR,IC,and so on.According to the Chinese Pharmacopoeia,the monitoring data is divided into several subsets based on drug category,which lays the data foundation for this study.2.Evaluation index:based on the existing grading（mild,moderate,severe）and scoring criteria of the damage of adverse drug reactions,combined with the“report type”（general or severe）and“adverse consequences”（death,sequelae,no improvement,improved,cured）field information is quantified for each ADR,and a signal risk evaluation indicator RI_ADR（Risk Indicator of ADR）is constructed,which is used to evaluate the pros and cons of the outlier signal detection results in this paper.3.Experimental simulation:the experiment in this paper is divided into three parts:（1）Overall inspection:A distance-based outlier detection method ODABD（Outlier Detection Algorithm Based on Distance）is proposed,which combines K-Means clustering algorithm to screen outliers in the overall data;（2）Signal detection based on data stratification:According to the Chinese Pharmacopoeia,the overall data is divided into 25 categories of subsets,and outlier detection is performed in each subset;（3）Signal detection based on context condition:Outliers is detected for each subset using“gender”as the context.Use the signal risk evaluation indicator RIADR to evaluate the results of the above three experiments.Experimental results show that the RIADR of the Context detection result is the largest.4.Results verification:Use information resources such as CFDA（China Food and Drug Administration,CFDA）,CNKI,and drug instruction to verify the results of the three outlier detection.The results show that the outlier detection method can screen out valuable signals that are obscured by conventional methods...