基于中医证候学的糖尿病肾病患者肾小球滤过率评估方法的探索

Exploration of GFR evaluation method in patients with diabetic nephropathy based on the theory of TCM syndrome

目的探索糖尿病肾病患者基于中医证候学评估肾小球滤过率(glomerular filtration rate,GFR)的可能性及其方法。方法基于"十一五"国家科技支撑计划《中医全程干预糖尿病肾病进程综合方案研究》的1872例研究数据,采用散点图矩阵、安德鲁斯曲线分析、平行坐标图等计算机可视化技术,探求GFR与性别、年龄、身高、体重,以及气虚、血虚、阴虚、阳虚、血瘀、湿浊、痰湿的中医证候积分等11个因素间的关系。将病例分为1400例的组1和472例的组2。基于组1的数据,使用两种方法进行GFR的估算:(1)对11个因素进行线性回归,并根据回归结果进行GFR估算。(2)以病例为单位,纳入11个因素,使用组1病例建立数据库,使用K最邻近结点算法(k-nearest neighbor,KNN),进行GFR估算。基于组2的数据,采用散点图、偏差分析、Bland-Altman作图法及ROC曲线进行验证一致性评价。结果 (1)GFR与性别、年龄、身高、体重以及7个中医证候等因素之间存在特定联系。(2)散点图显示,KNN法分布于±30%范围内的点较之回归方程明显增多。两种评估方法偏差的30%符合率均达到50%以上。回归方程和KNN法偏差的30%符合率分别达到58.1%和69.3%。Bland-Altman作图显示,KNN法估算结果的偏差分布较为均匀,数据相对集中,波动范围小于回归方程估算值。用于诊断肾功能不全时,KNN法的ROC曲线下面积达到0.847。结论基于中医证候学对糖尿病肾病患者的GFR进行评估是可行的。KNN法效果优于回归法,在数据量足够大时,更有利于中医证候学的研究。

Objective To explore the possibility of determining glomerular filtration rate（ GFR）in diabetic nephropathy（ DN） patients by TCM Syndrome,and the corresponding method. Methods Visualization technique was applied treating data from previous studies to explore the correlations between GFR and the 11 factors including gender,age,height,weight,and the TCM syndrome score for qi deficiency,blood deficiency,yin deficiency,yang deficiency,blood stasis,dampness,and phlegm dampness syndromes. The patients were divided into 2 groups with 1400 and 472 cases,respectively. 2 methods were demonstrated using group A data to evaluate GFR：（ 1） Linear regression GFR predictions by the aforementioned 11 factors.（ 2） K nearest neighbor（ KNN） predictions using the group A dataset. The prediction ef-fectiveness was evaluated by comparing the predicted results of group B data to the real ratio,and the evaluation is facilitated by scatter diagram,deviation analysis,Bland-Altman method and ROC curve. Results（ 1） GFR correlated to gender,age,height,weight and 7 of the TCM syndromes.（ 2） The scatter diagram shows that,in the range of ± 30%,the points of KNN is significantly increased compared with the regression equation. The deviation of 30% coincidence rate of both two kinds of assessment methods have reached more than 50%. The coincidence rate of deviation of 30% of linear regression and KNN method reached 58. 1% and 69. 3%. Bland-Altman mapping shows that,the deviation of the estimated results of KNN method is more evenly distributed. When the data is more concentrated,the fluctuation rage fell below the estimated value of the regression equation. The area of ROC curve of KNN method had reached 0.847 for the diagnosis of renal insufficiency. Conclusion The feasibility of predicting GFR by TCM syndromes is validated. The KNN method produces more robust results compared to linear regression. The amount of data may positively relate to a satisfactory TCM syndrome study.