基于决策树法探析高颖教授辨治失眠主方主症规律

迟辉 高颖

摘要 目的：探索高颖教授辨治失眠主方的关键适用症状应用规律，为临床诊疗提供参考。方法：采集高颖教授2015年5月至2020年1月门诊病例数据库中204例（349诊次）失眠患者的诊疗信息，采用SPSS Modeler软件进行决策树分析，用CHAID与CART算法构建主方关键适用症状的分类模型。结果：共建立丹栀逍遥散类方、一贯煎类方、小柴胡汤类方、柴芩温胆汤类方4个主方主症的分类模型，各模型的准确率分别为88.24%、90.68%、91.18%、94.61%，各模型的曲线下面积（AUC）值分别为0.87、0.92、0.89、0.90;共筛选出21个重要预测变量;共得到22项判断规则。结论：通过决策树法建立的分类模型能够有效反映高颖教授辨治失眠的主方主症规律，具有较高的准确度、灵敏度和特异度，可为失眠的临床诊治提供参考和指导。

关键词 失眠;辨证论治; @ 高颖;决策树;数据挖掘

Analysis on the Law of Professor GAO Ying′s Syndrome Differentiation and Treatment of Insomnia Based on Decision Tree

CHI Hui，GAO Ying

（Dongzhimen Hospital of Beijing University of Chinese Medicine，Beijing 100700，China）

Abstract Objective：To explore the key applicable symptoms of Professor GAO Ying′s main prescription for insomnia，and provide reference for clinical diagnosis and treatment.Methods：We collected the diagnosis and treatment information of 204 insomniac patients（349 consultations） in the physician′s outpatient case database in the past 5 years，using SPSS Modeler software to make a decision tree analysis，and used the CHAID and CART algorithm to build a classified model of the key applicable symptoms.Results：A total of 4 main models for the classification of the main symptoms of Danzhi Xiaoyao San Recipe，Yiguan Jianlei Recipe，Xiao Chaihu Tang Recipe，Chaiqin Wendan Decoction Recipe were established，and the accuracy rates of each model were 88.24%，90.68%，91.18%，94.61%.The AUC values of each model were 0.87，0.92，0.89，0.90; 12 important predictors were screened out，and 22 judgment rules were obtained.Conclusion：The classification model established by the decision tree method can effectively reflect Professor GAO Ying′s principle of treating insomnia with high accuracy，sensitivity and specificity.It can provide reference and guidance for the clinical diagnosis and treatment of insomnia.

Keywords Insomnia; Syndrome differentiation; @ GAO Ying; Decision tree analysis; Data mining

中圖分类号：R242;R749文献标识码：Adoi：10.3969/j.issn.1673-7202.2021.16.022

失眠在中医学称之为“不寐”，是当今社会的常见病症，严重影响人们的学习、工作和生命质量。既往研究多集中在失眠药物配伍规律，某类方剂疗效观察以及医师个体经验方药，缺乏对方剂或药物核心适用症状的精准分析[1-9]。决策树法常用于归纳名老中医辨证过程、挖掘中医临证诊疗规律[10]。由于其产生的分类规则简单直观，分类准确率较高，因此该方法适用于方药适用症状的精准分析。决策树是树形结构的模型，具有清晰的属性结构和较好的分类、预测能力，属于有监督学习中的一种，广泛使用于分类与预测任务中[11]。当目标变量是分类变量时，决策树完成分类任务;当目标变量为连续变量时，决策树完成预测任务。……