基于整合数据分析方法的糖尿病中西医结合适应性治疗策略研究

Diabetes is a chronic non-communicable diseases (NCD) and has many complications as well. While a diabetes patient is treated, a physician has to adjust treatment strategies based on individual factors and the special stage of disease processing. However, the treatment strategies are hard to be determined and proved because the evaluation of short-term effect and the optimization of treatment strategy are still challenges. The adaptive treatment strategies (ATS) could be an ideal method for these problems. In this study we would take diabetes treated by Traditional Chinese and Western Medicine as a research subject. First, we will screen 20 thousands clinical records and transform them into a structured database by structuring index system, and make 3 thousands research cases homogeneous. Then, we will use integrative data analysis (IDA) method to find how we integrate heterogeneous data sets, measure separated and lag effect of comprehensive intervention, and set up a syndrome-treatment-effect data network. Finally, we will mine individual adaptive treatment strategies for different types of diabetes patients treated by integrated traditional Chinese and western medicine. In this study, we will establish a suit of methods to judge ATS of diabetes. These methods will follow the principle that syndrome, treatment and effect should be put together closely in Traditional Chinese Medicine (TCM) theories. These methods will show us a path how we resolve some key issues in which treatment protocols are made based on individuals and adjusted dynamically, and physicians can consider applying these methods while they treat diabetes patients.

糖尿病是一种慢性疾病且合并症众多，在中西医结合治疗糖尿病过程中需要根据患者个体特征和疾病演变来不断调整治疗方案以达到最优疗效。但是，基于患者个体差异和疾病不同阶段最适宜的治疗方案难以确证，原因在于难以进行阶段性疗效评估及优选后续治疗方案，而“适应性治疗策略”研究可以解决该问题。本课题以中西医结合治疗糖尿病为例，选取2万份糖尿病电子病历数据进行结构化标引和规范化处理，对3千份科研病例数据进行归一化处理；然后基于整合数据分析方法探索异质异构数据的整合方法、糖尿病复杂干预的独立效应和延后效应测量方法，并在此基础上建立“证-治-效”数据网络；最后挖掘糖尿病的个体化中西医结合适应性治疗策略。本研究将创新性地建立个体化的糖尿病适应性治疗策略判定方法，体现中医证、治、效紧密相关的疗效评价原则，解决中西医结合治疗糖尿病过程中辨证论治、不断调整和优化治疗方案与动态效果评估的关键科学问题，为临床实践提供参考。

目的：糖尿病的治疗需要根据患者个体特征和疾病演变来不断调整治疗方案以达到最佳疗效。利用真实世界电子病历挖掘其“病、证、治”的组成规律、时序演变特点和诊疗策略，为提高糖尿病中西医结合治疗效果提供证据。.方法：从中国中医科学院中医药数据中心、深圳市中医院和开封市中医院采集回顾性2型糖尿病（T2DM）患者21127人（41497诊次）和923人（1105诊次）糖尿病肾病（DKD）患者。基于SNOMED和临床实际数据建立糖尿病临床术语规范化体系，构建医学短文本规范化模型；构建层次骨干发现模型（IHDF）以提取T2DM的合并疾病、证候、症状、治法、中药配伍的核心信息；构建多通道卷积神经网络模型对糖尿病“病、证、治”关系和“证、治”时序演变规律进行挖掘。.结果：（1）构建糖尿病17个一级，69个二级分类症状术语体系；（2）T2DM“病、证、治”网络结构主要呈幂律分布，高血压是最常见的合并疾病，病位主要在脾胃，病性以热为主，治疗以清热药为主；（3）T2DM证候可分为7个社团，表达原始信息90.67%，首要社团以虚和瘀为主；（4）T2DM证候时序变化较为稳定（51.37~66.64%），但长期会向虚或瘀方向转化；（5）T2DM治法可分为5个社团，表达原始信息100%，首要社团以滋阴清热和祛湿化痰为主；（6）T2DM证候早期患者以热为核心，在长期时序演变中发生证候变化者有22.32%，活血类中药使用增加；（7）DKD以本虚标实为主，血脂相关指标可影响证候分型；（8）语义相似度模型可实现单蕴涵文本86.18%准确识别；（9）IHDF算法在任意显著性水平下，均可提升复杂网络信息提取率；（10）多通道卷积神经网络模型在中药组方决策中F1值为51.2%，较对比算法提升约5.31%；（11）基于“君臣佐使”的处方识别算法F1值为77.99%，较对比算法提升11.04%。.结论：T2DM患者“病、证、治”关系存在明显规律，可以基于临床病历数据构建出T2DM中西医结合诊疗决策辅助模型；应关注T2DM疾病长期演变，掌握疾病变化规律以辅助制定决策；应根据中医临床数据特点制定适合的数据分析与挖掘算法。

内容获取失败，请点击重试