基于中医传统理论建立中药肾毒性的Logistic回归预测模型

Establishment of Logistic Regression Prediction Model for Nephrotoxicity of Chinese Herbal Medicine Based on Traditional Chinese Medical Theory

目的研究四性、五味和归经与中药肾毒性的相关性,建立中药肾毒性Logistic回归预测模型,并对模型进行评估。方法文献检索肾毒性中药111味、非肾毒性中药398味。记录上述中药的四性(寒、热、温、凉和平)、五味(酸、苦、甘、辛、咸、淡和涩)和归经(肺经、大肠经、胃经、脾经、心经、小肠经、膀胱经、肾经、心包经、三焦经、胆经和肝经),进行肾毒性中药、非肾毒性中药与其四性、五味和归经的相关性检验,相关变量进行非条件Logistic回归,并建立Logistic回归预测模型,绘制Logistic回归预测模型概率的ROC曲线,评价预测模型。结果肾毒性中药和非肾毒性中药在四性归属有相关性(P<0.05),相关系数为0.241;肾毒性中药和非肾毒性中药在五味归属有相关性(P<0.05),相关系数0.210;肾毒性中药和非肾毒性中药在归经归属无相关性(P>0.05)。四性、五味两个方面12个变量因素入选Logistic回归分析,筛选出四性中的热、平和五味中的苦、甘共4个变量因素,建立Logistic回归预测模型(预测概率分界值为0.28)。结论中药四性、五味、归经3个方面24个变量因素当中,四性、五味两个方面12个变量因素与其肾毒性具有相关性,其中四性中的热、平和五味中的苦、甘为肾毒性的影响变量。在此基础上建立的Logistic回归预测模型对于中药肾毒性具有一定的预测能力。

Objective To study the correlation of nephrotoxicity of Chinese herbal medicine with the four properties,five flavors and channel tropism for the establishment of prediction model of nephrotoxicity,and to evaluate the model.Methods The data of 111 kinds of Chinese herbal medicine with nephrotoxicity and 398 kinds of Chinese herbal medicine without nephrotoxicity were collected after literature search. The collected data covered the four properties（cold,hot,warm,cool,and neutral）,five flavors（sour,bitter,sweet,pungent,salty,tasteless and astringent）and channel tropism（lung meridian,large intestine meridian,stomach meridian,spleen meridian,heart meridian,small intestine meridian, bladder meridian, kidney meridian, pericardium meridian, triple energizer meridian, gall bladder meridian,and liver meridian）of the included herbs. The correlation of nephrotoxic or non-nephrotoxic Chinese herbal medicine with the four properties,five flavors and channel tropism was analyzed. The unconditional Logistic regression（LR） was applied to screen the variables to be input. LR prediction model was set up. Receiver operating characteristics（ROC）curve of the LR prediction model was drawn,the area under the curve,sensitivity,specificity and overall accuracy were calculated,and the accuracy of prediction model was evaluated. Results Chinese herbal medicine with or without nephrotoxicity was correlated with the four properties and five flavors（P 0.05）,with the coefficient being 0.241,0.210 respectively. However,Chinese herbal medicine with or without nephrotoxicity was not correlated with the channel tropism（P 0.05）. There were 12 variables selected for the LR from four properties and five flavors. Four variables, which were hot and neutral of the properties, bitter and sweet of the flavors, were selected to establish the LR prediction model（cutoff value being 0.28）. Conclusion The 12 of 24 variables from four properties, five flavors and channel tropism of Chinese herbal medicine, are correlated with the nephrotoxicity of Chinese herbal medicine,of which 4 variables including hot and neutral from four properties,bitter and sweet from five flavors affect the nephrotoxicity. Based on the results,the established LR prediction model has some predictive value for nephrotoxicity of Chinese herbal medicine.