OBJECTIVE: To establish an unbalanced data processing method to re-evaluate the clinical application safety of Danhong injection（DHI）

to predict the occurrence factors of clinical adverse drug reactions

so as to provide reference for the rational drug use in clinic. METHODS: Firstly

a “multi-center and large sample” hospital-based surveillance method was used to collect 30 888 inpatients who used DHI from Apr. 2009 to Aug. 2013 in 37 general hospitals in 6 provinces across China

and 108 adverse drug reaction cases were identified

with an incidence of 3.5‰

which belonged to unbalanced data. Based on the unbalanced data processing method

the resampling-regularization method was constructed to analyze and process patients’ basic information

symptoms

drug administration and drug combination to determine the factors associated with adverse drug reactions. RESULTS: Based on the constructed unbalanced data processing method

six resampling-regularization models were selected

and the combination of random under-sampling and synthetic minority over-sampling technique（in the ratio of 3∶ 3） was found to be a better resampling model. A total of 41 DHI adverse drug reaction-related variables were obtained by using no less than three model selections as screening conditions. Analysis showed that women

patients with a lower body mass index and with a history of allergies were more likely to have adverse drug reactions. Patients using DHI for the first time were more likely to have adverse drug reactions

and drug combination was prone to cause adverse drug reactions

yet the combination with spironolactone

aspirin

ambroxol hydrochloride and benazepril hydrochloride might alleviate the occurrence of adverse drug reactions caused by DHI. CONCLUSIONS: The constructed resampling-regularization strategy can be used to deal with the unbalanced data of adverse drug reactions caused by DHI

and then perform post-marketing safety re-evaluation and accurate prediction of risk factors to guide rational clinical use and drug combination warning

yet whether the constructed algorithm can be used for post-marketing re-evaluation of other proprietary Chinese medicines needs further study.