【摘要】 【目的】基于机器学习方法构建包含中医证候的特发性膜性肾病（IMN）患者的疾病进展风险预测模型，以期量化评价中医证候在IMN的疾病进展风险预测中的价值。【方法】利用单因素分析、递归消除法（RFE）和多因素二元Logistic回归分析方法筛选影响IMN的疾病进展风险的独立相关因素，并构建风险预测模型。将102例IMN患者按65∶35的比例随机分配至训练集和测试集，比较纳入或不纳入证候信息特征的风险预测模型的性能指标如精确度、敏感度、特异性、F1值和受试者工作特征（ROC）曲线下面积（AUC）的变化。【结果】未纳入证候信息特征之前，经单因素分析结合RFE筛选得到IMN患者年龄、血红蛋白定量、尿潜血、24 h尿蛋白定量、尿蛋白肌酐比、肾小球滤过率（eGFR）、肌酐、尿酸、谷丙转氨酶、抗磷脂酶A2受体抗体（PLA2R-Ab）、总胆固醇和低密度脂蛋白胆固醇共12个临床特征变量。构建含有上述变量的风险预测模型，经多因素二元Logistic回归分析后，结果显示训练组和测试组中以上临床变量均具有统计学意义，且该风险预测模型具有良好的敏感性和预测性。将证候信息特征纳入后再次运用RFE法，筛选出14个特征变量，其中血瘀证和湿阻证被纳入，结果显示风险预测模型的敏感度、特异性等指标较未纳入证候信息特征前有了明显的提高。【结论】研究结果初步表明中医证候是IMN的疾病进展风险预测重要的补充特征，可为今后中西医信息联合的智能化诊断提供参考，为后续的中医药治疗起到指导作用。更多还原

【Abstract】 Objective To construct a model combining with traditional Chinese medicine(TCM)syndrome for predicting the risk of disease progression in patients with idiopathic membranous nephropathy(IMN)by machine learning methods, thus to quantitatively evaluating the value of TCM syndrome in the prediction of the risk of disease progression in IMN. Methods Monofactor analysis,recursive feature elimination(RFE)and multivariate binary Logistic regression analysis were used to screen the independent related factors affecting the risk of disease progression of IMN, and then a risk prediction model was constructed. A total of 102 patients with IMN were randomly assigned to the training set and the test set in a ratio of 65∶35,and then the comparison was conducted in the performance indicators of accuracy,sensitivity,specificity,F1 value,and area under the receiver operating characteristic(ROC)area under the curve(AUC)of the risk prediction model with or without the inclusion of the TCM syndrome information. Results Before the inclusion of TCM syndrome information,12 clinical characteristic variables for patients with MN were obtained after monofactor analysis combined with RFE screening, and they were age,hemoglobin quantification,urinary occult blood,24-hour urine protein quantification,urine proteincreatinine ratio,estimated glomerular filtration rate(eGFR),creatinine,uric acid,alanine transaminase,antiphospholipase A2 receptor antibody(PLA2R-Ab),total cholesterol,and low-density lipoprotein cholesterd. A risk cholesterol prediction model containing the above variables was constructed. The multivariate binary Logistic regression analysis showed that the differences of the clinical variables mentioned above between the training-set group and test-set group were statistically significant,and the risk prediction model presented good sensitivity and predictability. Monofactor analysis combined with RFE screening was performed again after the inclusion of TCM syndrome information,and then 14 variables were obtained,which included blood stasis syndrome and dampness obstruction syndrome. The sensitivity and specificity of the model with the inclusion of the TCM syndrome information were significantly improved when compared with those without the inclusion of TCM syndrome information. Conclusion The results of the study initially indicate that TCM syndrome can be used as an important supplementary variable for predicting the risk of disease progression in IMN, and will provide a reference for intelligent diagnosis through the integration of traditional Chinese and western medicine information,and will supply the guidance for the treatment of IMN with TCM.更多还原