摘要
目的:应用多元统计方法对胃食管反流病(gastroesophageal reflux disease,GERD)郁热阴伤证进行探讨研究,为GERD中医证候的分类和诊断标准的制定提供一定的客观依据.方法:选取402例GERD患者,制定临床证候信息采集表,采集相关信息,建立中医证素数据库,随机分为训练样本组和验证样本组.以训练样本组数据建立相关量化诊断模型,采用二元logistic逐步回归分析进行相关因素筛选,运用CART决策树对诊断模型进行剪枝,运用条件概率换算公式建立相关因素赋分表,通过绘制ROC曲线选择最佳诊断阈值,根据诊断性试验评价原则,分别计算训练样本组和验证样本组数据的灵敏度、特异度、准确度、阳性似然比,对量化诊断模型进行回顾性研究和前瞻性检验.结果:共提取28项中医证素,经过逐步回归分析后得到16项相关因素,其中包括正相关因素8项,负相关因素8项,进一步运用CART决策树精简后得出6项中医证素,筛选出2条判断规则.训练样本组和验证样本组ROC曲线下面积、检验灵敏度、特异度、准确度、阳性似然比分别为0.993、95.0％、97.0％、96.3％、32.0和0.949、87.5％、92.7％、90.1％、12.0.模型具有较好的科学性和实用性.结论:基于lo-gistic 回归分析、ROC曲线和CART决策树算法可以有效建立GERD郁热阴伤证诊断模型,为临床辨证论治提供了客观依据.
Abstract
Objective:To explore the Traditional Chinese Medicine(TCM)diagnostic features of the syndrome of stagnated heat and yin injury in gastroesophageal reflux disease(GERD),by using the multivariate statistical methods,so as to provide some objective basis for the formulation of diagnostic criteria.Methods:The TCM syn-drome data of 402 GERD patients were collected to develop a clinical symptom information collection table,collect relevant information,and establish a TCM syndrome database which were randomly divided into the training sam-ple group and verification sample group.A quantitative diagnosis model was established based on the training sam-ple group data,the binary logistic stepwise regression analysis was used to screen related factors,conditional probability conversion formulas were used to establish a score table for related factors,the CART decision tree is used to prune the diagnostic model,ROC curves was drawn to select the best diagnostic threshold,the sensitivity,specificity,accuracy,and positive likelihood ratio were calculated respectively according to the diagnostic test eval-uation principles,the quantitative diagnostic model was analyzed and evaluated,and finally the validation sample group data was used to complete the scientific attribute evaluation of the model.Results:A total of 28 clinical fea-tures were extracted.After stepwise regression analysis,16 relevant factors were finally obtained,8 positively correlated factors,8 negatively correlated factors;then we got six TCM syndrome elements by the CART,which screening out two judgment rules;after quantitative scoring.The ROC curve,sensitivity,specificity,accuracy,and positive likelihood ratio of retrospective test and prospective test were 0.993,95.0％,97.0％,96.3％,32.0 and 0.949,87.5％,92.7％,90.1％,12.0 respectively.The model had good scientificity and practicality.Con-clusion:Based on Logistic regression analysis,ROC curve and CART decision tree algorithm,the diagnostic mod-el of gastroesophageal reflux heat-yin injury syndrome can be effectively established,which provides an objective basis for clinical syndrome differentiation and treatment.
维普
万方数据