Real-world Based Safety Evaluation Of Guhong Injection

Objective: With the increasing of clinical dosage of Guhong Injection,the reported cases of adverse drug reaction（ADR）also gradually increased.However,the risk factors affecting the occurrence of adverse reactions was still unclear.By collecting ADR case reports of Guhong Injection in the real-world and using data mining methods,thesis explored the potential risk factors of ADR and established Guhong injection ADR prediction model,to provide basis for safe and rational use of Guhong injection in clinic.Methods: 304 effective ADR reports of Guhong injection received by the drug SRS of Sichuan ADR monitoring center from January 1,2019 to August 30,2021 were collected.After data cleaning,the Apriori algorithm was used for correlation analysis.The ADR cases collected in the real-world were matched according to the Nested CaseControl Study 1:4,and then SPSS modeler was used for data mining to establish adverse reaction prediction model and evaluate the model.Result: Through the correlation analysis of ADR reports of Guhong injection collected in SRS,the main population was 60-79 years old,and there were more women in terms of gender.The adverse reactions mainly occured in the skin and accessory injuries,cardiovascular system injuries and systemic injuries.Finally,random forest（RF）,support vector machine（SVM）and logistic regression（LR）were selected for data mining.The three models could achieve 100% accuracy in the training set,while the RF model had the highest prediction accuracy in the test set,reaching 96.3%.From the receiver operating characteristic curve（AUC）,the RF model and SVM model were0.986 and 0.915 respectively,while the AUC value of LR model was only 0.77.In terms of recall rate,RF model performs best,with a recall rate of 81.8%,SVM model with a recall rate of 63.6%,while the LR model had only 54.5%.Conclusion: The occurrence time and performance of ADR of Guhong injection were affected by the basic information of patients and diseases.The risk factors for ADR were single dose,age of patients,combined use of anti-arrythmia drugs and anticholinergic drugs.Among the three models,RF had high prediction accuracy and the best generalization ability,which could be used to achieve clinical value.