基于ROC曲线的中药致敏风险预测模型研究  

Predicting Anaphylaxis of Chinese Herbal Medicine—A Roc Curve Analysis

作　　者：谢晴宇[1] 孟庆刚[2] 王忠[1] XIE Qingyu;MENG Qinggang;WANG Zhong(Institute of Basic Research in Clinical Medicine,China Academy of Chinese Medical Sciences,Beijing 100700,China;Beijing University of Chinese Medicine,Beijing 100029,China)

机构地区：[1]中国中医科学院中医临床基础医学研究所,北京100700 [2]北京中医药大学,北京100029

出　　处：《中华中医药学刊》2018年第12期3005-3009,共5页Chinese Archives of Traditional Chinese Medicine

基　　金：国家自然科学基金项目(81674102);中国博士后基金项目(2014M550133)

摘　　要：目的:探索中药成分致敏性与总脱靶数量及致敏脱靶数量的关系。方法:参考《按照传统既是食品又是中药材物质目录管理办法》的中药形成“安全中药成分数据集”。参考CFDA发布的药品不良反应信息通报,确定有致敏风险的中成药品种,作为“致敏风险中成药数据集”。对两组中药成分进行抽样和脱靶预测(主要预测总脱靶数量及致敏脱靶数量两项指标),其后,进行ROC曲线分析,探索3种模型预测成分致敏风险的最优临界值(cutoff)。结果:致敏成分组与安全成分组总脱靶数量和致敏脱靶数量相比均具有统计学差异(P<0.01)。结论:由总脱靶个数和致敏脱靶个数共同建立的综合预测指标具有最好的预测准确度(AUC=0.81)。Objective: To explore the association between the hypersensitivity of molecules in Chinese medicine and their total off-targets(TO) and/or hypersensitivity off-targets(HO). Method: Safety herbs which were simultaneously edibles and Chinese medicine(abbreviated as edible-medicinal herbs) were identified referring to materials such as regulation on the list of traditional edible-medicinal Chinese herbs. Chinese medicinal products were potentially hypersensitive according to CFDA alerts. After duplication removal, the rest products were pooled to generate Chinese medicine products with hypersensitive risk dataset. Predictions on the number of TO and HO of substances in NHG and HG were made and recorded. ROC analysis was performed to generate 3 models with TO, HO and prediction scores of binary logistic regression combining TO and HO to identify the optimal cutoff for the 3 models. Result: Significant differences are identified between NHG and HG(P<0.01). Conclusion: Best prediction performance is observed using TO+HO model(AUC=0.81).

关 键 词：致敏预测 药食同源 CFDA药物不良反应警戒 ROC曲线 

分 类 号：R2-03[医药卫生—中医学]