基于中医四诊大数据的冠心病风险评估与预测模型研究

The project carries out the research from retrospective and prospective aspects. Based on the database of objective parameters of inquiry scale, tongue, face, pulse-taking and listening in traditional Chinese medicine (TCM) and clinical physical and chemical indicators: ①Establish the diagnostic model of the disease and syndromes in CHD based on multi label learning method, deep learning, complex system method and other data mining methods. ②Screen the risk factors for CHD by analyzing the factors such as general condition of the patients (including body mass index, drinking, smoking, high salt diet, family history), physical and chemical indexes (including blood pressure, blood glucose, blood lipid, coronary angiography, echocardiography), TCM four diagnostic information (including inquiry information, parameters of tongue and face, parameters of pulse, parameters of auscultation, and palm information) and TCM syndrome (type of constitution of patients, disease syndrome differentiation) with CHD incidence. ③From the aspects of multi-approach and multi-level, analyze the occurrence of CHD and the risk factors of the risk events resulted from CHD, such as Heart, brain, kidneys and other risk events; research on the correlation between parameters of four diagnostic methods in TCM and the occurrence of CHD and the risk factors of the risk events; establish the models of the risk assessment and prediction for CHD. Evaluate the models described above. Provide the inexpensive and effective methods for the prevention of the occurrence of CHD and function injury of target organs; provide the scientific basis for the application of the detecting instrument of four diagnostic methods in TCM on research and clinical study.

本项目从回顾性和前瞻性两方面进行研究。基于中医问诊量表和舌、面、脉、声诊的客观参数及临床理化检测指标的数据库：①基于多标记学习方法、深度学习等多种数据挖掘方法及复杂系统方法，建立冠心病病证诊断模型。②分析患者一般情况（包括体质量指数、饮酒、吸烟、高盐饮食、家族史）、理化检测指标（包括血压、血糖、血脂、冠脉造影、心脏彩超）、中医四诊信息（包括问诊信息、舌象参数、脉象参数、声诊参数和掌纹信息）及中医证候（患者的体质类型、疾病的辨证分型）等因素对冠心病发病的影响，筛选出冠心病危险因素。③从多途径、多层次对冠心病发病及其引起心、脑、肾等风险事件的危险因素进行分析，探讨中医四诊客观参数与冠心病发病及其风险事件的相关性，建立冠心病风险评估与预测模型，对模型进行评价。以期提供经济有效的预防冠心病的发生、预防靶器官功能损伤的方法，为中医四诊检测仪器在临床和科研中的推广应用提供科学依据。

本项目从回顾性和前瞻性两方面进行研究。建立了基于中医问诊量表和舌、面、脉、声诊的客观参数及临床理化检测指标的数据库。在此基础上：①利用809例冠心病患者的中医四诊信息，基于多粒度级联森林算法建立了冠心病中医证候诊断模型，对心气虚证、心阳虚证、心阴虚证、心血虚证、痰浊证、血瘀证的证型分类准确率分别达到了85.2%、91.3%、81.3%、97.9%、64.1%和63.4%。②通过分析冠心病患者一般情况（包括体质量指数、饮酒、吸烟、高盐饮食、家族史）、理化检测指标（包括血压、血糖、血脂、冠脉造影、心脏彩超）、中医四诊信息（包括问诊信息、舌象参数、脉象参数、声诊参数和掌纹信息）及中医证候（患者的体质类型、疾病的辨证分型）等因素，筛选出了一系列与冠心病发病和继发风险事件相关的指标参数，并在此基础上建立了基于CHAID决策树的冠心病患者风险分层识别模型，基于四诊参数的CHAID决策树模型对冠心病高危患者的识别准确率可达78.95%。③基于随机生存森林算法对530例随访冠心病患者进行生存分析，建立冠心病风险评估与预警模型，对冠心病患者发生心、脑、肾等事件的风险进行预测和评估，在测试集上的2年tdROC曲线下面积达0.809，以该模型为依据将测试集分为低风险组和高风险组，两组的Kaplan-Meier生存曲线有显著性差异。该模型可以提供有效的冠心病风险评估和预警手段，为中医四诊检测仪器在临床和科研中的推广应用提供科学依据。