慢性心力衰竭合并2型糖尿病患者易损期心源性死亡的中西医预警指标分析

Analysis of Chinese and Western warning indicators of cardiogenic death in chronic heart failure combined with type 2 diabetes mellitus during the vulnerable phase

目的:分析慢性心力衰竭合并2型糖尿病患者发生易损期心源性死亡的中西医预警指标,为临床实践提供循证医学证据。方法:将山东省7家中医院符合研究标准的727例患者纳入队列研究。运用Cox回归结合多重支持向量机-递归特征消除(MSVM-RFE)模型分析预警指标。结果:将Cox单因素回归分析获得的53个潜在预警指标按重要性排序依次纳入MSVM-RFE模型进行效能检验,当纳入前11位潜在预警指标时,获得最小错误率0.177与最大精度0.823,即为最优预警指标子集。Cox多因素回归分析显示,纽约心脏病学会心功能分级(Ⅲ级vsⅡ级:HR=2.33,Ⅳ级vsⅡ级:HR=2.59),吸烟史(HR=1.75),血浆氨基末端脑利钠肽前体(NT-ProBNP)(HR=8.08),气阴两虚血瘀证(HR=2.26),面色口唇紫暗(HR=3.63),脉细数无力或结代(HR=1.02),少苔或有裂纹(HR=1.05),空腹血糖(FBG)(HR=1.02)为损害性预警指标;益气类注射液(HR=0.44)、降糖类药物(HR=0.29),降血脂类药物(HR=0.96)为保护性预警指标。结论:Cox回归与MSVM-RFE模型相结合的数据分析方法可实现优势互补,适用于高维、多重共线性、小样本数据的中西医预警指标分析,具有临床推广价值。

Objective:To analyze the warning indicators of Chinese and Western medicine in patients with chronic heart failure combined with type 2 diabetes mellitus who develop cardiogenic death during the vulnerable phase,and to provide evidence-based medical evidence for clinical practice.Methods:A total of 727 patients from 7 hospitals in Shandong Province who met the study criteria were included in the cohort study.Cox regression combined with multiple support vector machine-recursive feature elimination(MSVM-RFE) model was used to analyze the warning indicators.Results:The 53 potential warning indicators obtained from the Cox univariate regression analysis were sequentially included in the MSVM-RFE model in order of importance for the effectiveness test,and when the top 11 potential warning indicators were included,the model obtained a minimum error rate of 0.177 and a maximum precision of 0.823,so they were used as the optimal subset of warning indicators.The Cox multifactorial regression analysis showed that the New York Heart Association(NYHA) cardiac function classification(Class Ⅲ vs Class Ⅱ:HR=2.33,Class Ⅳ vs Class Ⅱ:HR=2.59),smoking history(HR=1.75),plasma amino-terminal brain natriuretic peptide precursor(NT-ProBNP)(HR=8.08),qi and yin deficiency and blood stasis syndrome(HR=2.26),purple face and lips(HR=3.63),thin and weak pulse or knotted pulse(HR=1.02),fissured tongue with little coating(HR=1.05),fasting blood glucose(FBG)(HR=1.02)are damaging warning indicators.Yiqi injections,hypoglycemic drugs,and hypolipidemic drugs Yiqi injections(HR=0.44),hypoglycemic drugs(HR=0.29),hypolipidemic drugs(HR=0.96) are protective warning indicators.Conclusion:The combination of MSVM-RFE and Cox regression data analysis method could achieve complementary advantages.It is suitable for the analysis of Chinese and Western medicine warning indicators with high-dimensional,multi-collinearity and small sample data,and has clinical promotion value.