基于数据挖掘的亚健康多维特征研究

发布时间：2019-07-06 15:17

【摘要】：背景：世界卫生组织的一项全球性调查显示,真正健康的人仅占5%,患有疾病的人占20%,而75%的人处于健康和患病之间的过渡状态,世界卫生组织将其称之为“第三状态”,我国称之为“亚健康状态”。在2010年中国亚健康学术研讨会上,专家们预测中国目前处于亚健康状态的人群已达70%。而亚健康状态是疾病发生前的过渡阶段,如不及时加以干预,将有可能进一步发展为疾病,从而增加社会的医疗负担。可见亚健康已成为我国一个不容忽视的公共卫生问题。中医学以其“整体观念”“辨证论治”的思想,在亚健康状态的干预方面具有很大优势。随着亚健康研究的开展,亚健康的概念与诊断已较为规范,并形成了《亚健康中医临床指南》,规范了亚健康的概念及相关诊断方法,但对亚健康状态的干预尚未形成针对性、系统性的方法。笔者认为,这是由于亚健康状态的表现特征众多,包括不适症状、生活方式、中医体质、心理状态等多个方面,且不同的特征处于量变积累、相互影响的过程中,表现或多或少,或重或轻,这使到亚健康的特征不易把握,干预方向难以准确,故尚未形成针对性、系统性的方法,疗效也因此参差不齐。有鉴于此,加强亚健康状态的多维特征研究,对其特征进行分析、归纳显得非常必要。目的：通过横断面的流行病学调查,了解亚健康人群多维特征的分布情况。利用数据挖掘技术,结合临床知识,对亚健康多维特征进行简化、归纳,探索特征之间的纵向重要性关系和横向关联性关系等。这将有利于进一步认识亚健康特征,为制定系统性、针对性的干预方向提供思路。方法：采用国家十一五科技支撑计划《亚健康状态的中医辨识与分类研究》课题组研制的《个体身心健康调查量表》,于2008年5月-2009年5月对来自华东、华北、华南、西南9家合作单位(广东省中医院、炎黄东方(北京)健康科技有限公司、解放军总医院亚健康研究所、成都中医药大学附属医院、上海中医药大学附属龙华医院、中国中医科学院广安门医院、南京中医药大学附属医院、浙江省中医院、重庆市中医院)的体检人员进行健康状态辨识及多维特征的流行病学调查,建立亚健康状态多维特征数据库。首先,对亚健康状态多维特征进行描述性分析,形成第一级亚健康多维特征池。采用单因素分析,筛选出健康人群与亚健康人群的区别特征,形成第二级亚健康多维特征池。采用聚类分析、结合临床知识对第二级亚健康多维特征池的不适症状维度,生活方式维度进行简化、归纳,形成第三级亚健康多维特征池。然后,利用决策树模型,分别对三个不同级别的亚健康多维特征池进行建模,形成三个亚健康特征纵向关系模型,结合临床知识对三个决策树模型的诊断特性进行比较,选出较为优化的决策树模型和多维特征池。接着,利用关联规则模型,对较为优化的多维特征池进行分析,形成亚健康特征横向关系模型。最后,结合临床知识对上述两个模型进行综合的分析解读。结果：一共收集,符合纳入标准、不符合排除标准样本4086例,其中健康人群1716例(占49.13%),亚健康人群1777例(占50.87%)。依据文献研究结果,将亚健康多维特征归纳为六大维度,分别是不适症状特征、人口学特征、生活方式特征、中医体质特征、社会心理特征、生存质量特征。共收集六大维度特征228个,其中不适症状特征117个、人口学特征28个、生活方式特征38个、中医体质特征9个、社会心理特征27个、生存质量特征9个,形成第一级亚健康多维特征池。采用单因素分析,筛选出健康人群与亚健康人群的区别特征共197个,其中不适症状特征94个、人口学特征28个、生活方式特征33个、中医体质特征9个、社会心理特征25个、生存质量特征8个,形成第二级亚健康多维特征池。采用聚类分析、结合临床知识对第二级亚健康多维特征池的不适症状、生活方式维度进行简化归纳,共得出特征96个,其中不适症状特征13个,人口学特征22个,生活方式特征19个,中医体质特征9个,社会心理特征25个,生存质量特征8个,形成第三级亚健康多维特征池。利用决策树模型,对三个不同级别的亚健康多维特征池进行建模,第一级特征池决策树模型与第二级特征池决策树模型完全相同,其诊断符合率为72.5%,对于亚健康状态的灵敏度为75.8%,特异度为69.1%,阳性预测值为71.8%,阴性预测值为73.4%,ROC曲线下面积为0.72。第三级特征池决策树模型其诊断符合率为76.2%,对于亚健康状态的灵敏度为77.2%,特异度为75.1%,阳性预测值为76.2%,阴性预测值为76.1%,ROC曲线下面积为0.76。统计结果提示,三个特征池决策树模型相对于参考线均有诊断意义(P0.05),第三级特征池决策树模型的曲线下面积比第一、二级特征池决策树模型大,且存在统计学差异(P0.05),故选择较为优化的第三级特征池决策树模型作为亚健康特征纵向重要性关系模型。再利用关联规则模型,对较为优化的第三级亚健康多维特征池进行分析,形成亚健康特征横向关系模型。共得出关联规则142条,最低支持度为50.08%,最高支持度为85.20%,最低置信度为90.08%,最高置信度为95.57%,形成亚健康特征横向关联关系模型。结论：结果表明亚健康人群的多维特征众多,根据文献研究及流行病学调查结果,可分为不适症状特征、人口学特征、生活方式特征、中医体质特征、社会心理特征、生存质量特征,共六大维度。在临床知识基础上,结合单因素分析、聚类分析对亚健康多维特征进行简化、归纳,并利用决策树模型、关联规则模型对亚健康多维特征的纵向、横向关系进行探索,可以为临床干预亚健康状态提供证据与思路：亚健康人群的干预体系应以中医体质为中心开展,对于偏颇体质人群,可针对其偏颇体质进行干预,且偏颇质人群在干预体质的同时应注意加强良好的饮食习惯宣教与脾胃的调养。此外,偏颇体质人群还较容易合并人际关系敏感的心理异常,应注意及时排查,并结合职业特性进行疏导。对于平和质人群,虽然没有出现体质的偏颇,但容易出现气虚、阴虚的中医证候特征,可对其进行干预。并且应重视良好的饮食习惯宣教和结合人群所处的地域、性别特性。无论体质偏颇与否,血虚、肾膀胱的证候要素特征比较容易被兼夹,需要在干预时注意。
[Abstract]:Background: A global survey by the World Health Organization shows that the real health population accounts for only 5%, people with a disease account for 20%, while 75% are in a transitional state between health and illness, and the World Health Organization calls it a "the third state", which is called the "sub-health state" . At the 2010 Chinese sub-health academic seminar, experts predicted that the population currently in sub-health has reached 70 %. The sub-health status is the transitional phase before the disease occurs. If the intervention is not carried out in time, it is possible to further develop the disease as a disease, thus increasing the medical negative of the society The sub-health has become a public health question that can't be ignored in our country In the sub-health state, the concept and diagnosis of sub-health have been standardized and the sub-health Chinese medicine is formed. The guideline>, the concept of sub-health and the related diagnosis method, but the intervention of sub-health state has not formed a targeted and systematic way. In many aspects, such as the physical and mental state of the doctor, and the different characteristics are in the process of quantitative accumulation and mutual influence, it is more or less, heavy or light, which makes it difficult to grasp the characteristics of sub-health, and the intervention direction is difficult to be accurate, so it has not formed the pertinence and systematicness. In view of the above, the multi-dimensional character of sub-health state is strengthened, and its characteristics are analyzed, and it is concluded that it is not It is often necessary . Purpose: cross-section According to the epidemiological investigation, the distribution of the multi-dimensional characteristics of the sub-health population is known. The data mining technology is combined with the clinical knowledge to simplify the sub-health multi-dimensional characteristics, to sum up, to explore the relationship between the longitudinal importance and the cross-correlation between the features, etc. This will be beneficial to the further To recognize sub-health characteristics and to make a systematic and targeted work predirection Methods: to adopt the national five-year scientific and technological support plan the study of the individual physical and mental health questionnaire developed by the research group, and May-2,2008 In May,009,9 cooperation units from East China, North China, South China, and Southwest China (Guangdong Central Hospital, Yanhuang Dongfang (Beijing) Health and Technology Co., Ltd., Sub-health Research Institute of PLA General Hospital, Affiliated Hospital of Chengdu University of Traditional Chinese Medicine and Shanghai The health status identification and the epidemiological investigation of the multi-dimensional characteristics of the physical examination personnel of the Affiliated Hospital of the Chinese Medical University, the Guangan Men Hospital of the Chinese Academy of Chinese Medicine, the Affiliated Hospital of Nanjing University of Traditional Chinese Medicine, the Central Hospital of Zhejiang Province and the Central Hospital of Chongqing, and the establishment of sub-health status First, a descriptive analysis of the multi-dimensional characteristics of sub-health state is carried out to form a first-order sub-health multi-dimensional feature pool, The method comprises the following steps: using cluster analysis, combining clinical knowledge to the discomfort symptom dimension and the living mode dimension of the second-level sub-health multi-dimensional feature pool, and inducing and forming a third-level sub-sub-health multi-dimensional feature pool; then, using the decision tree model, the sub-health multi-dimensional characteristic pool is respectively used for three different levels of sub-sub-health multi-dimensional characteristic pool, The health multi-dimensional characteristic pool is modeled to form three sub-sub-health characteristic longitudinal relationship models, the diagnosis characteristics of the three decision tree models are compared with the clinical knowledge, and the optimized decision tree model and the multi-dimensional characteristic pool are selected. analyzing the multi-dimensional characteristic pool to form a sub-health characteristic transverse relation model, and finally, combining the clinical knowledge to the two model entry Analysis and interpretation of the comprehensive analysis of the line. Results: The total collection, compliance with the inclusion criteria, and the exclusion criteria were not met in 4086 cases, of which the healthy person 1716 cases (49.13%) and 1777 (50.87%) of sub-health population were divided into six dimensions according to the results of the literature. The characteristics of physical characteristics, social and psychological characteristics and quality of life of the Chinese medicine were collected. The characteristics of six dimensions were collected. Among them, there were 117 symptoms,28 demographic characteristics,38 life styles,9 physical features,9 social cores. The first-stage sub-sub-health multi-dimensional characteristic pool was formed by using the single-factor analysis, and the differences between the healthy population and the sub-health population were analyzed. Among them, there were 94 non-appropriate symptoms,28 demographic characteristics,33 life-style features, and the traditional Chinese medicine. The characteristics of 9,25 and 8 life-quality features of the second-level sub-health multi-dimensional feature pool were formed by cluster analysis and combined with the clinical knowledge, and the characteristics of discomfort and life style of the second-level sub-health multi-dimensional feature pool were simplified and summarized, and the characteristics of the two-level sub-health multi-dimensional characteristic pool were obtained. 96 of them, including 13 of the symptoms,22 in the demography,19 in the lifestyle,9 in the constitution of the Chinese medicine,25 in the social psychology and 8 in the quality of life. and forming a third-stage sub-health multi-dimensional feature pool, The first-level feature pool decision tree model is the same as the decision tree model of the second-level feature pool, the diagnostic coincidence rate of the first-level feature pool is 72.5%, the sensitivity to sub-health state is 75.8%, the specificity is 69.1%, and the positive predictive value is 71.8%, the negative predictive value was 73.4%, and the area under the ROC curve was 0.72. The diagnostic accuracy of the decision tree model was 76.2%, 77.2% for sub-health status, 75.1% for specificity, 76.2% for positive predictive value and 76.1% for negative predictive value. The area of the curve under the curve is 0.76. The statistical results indicate that the decision tree model of the three feature pools has a diagnostic significance with respect to the reference line (P0.05). The lower area of the decision tree model of the third-level feature pool is lower than the first and the second-order characteristic pool decision tree model. At the same time, there is a statistical difference (P0.05). Therefore, the decision tree model of the third-level characteristic pool is selected as the model of the longitudinal importance of sub-health. The third-level sub-health multi-dimensional feature pool is analyzed to form a sub-health feature transverse relationship model. The association rule 142 is obtained. The minimum degree of support is 50.08%, the highest degree of support is 85.20%, the minimum confidence is 90.08%, and the highest confidence is 95.57. %, shape Conclusion: The results show that the multi-dimensional characteristics of sub-health population are many. According to the study of the literature and the epidemiological survey, it can be divided into the characteristics of discomfort, the characteristics of demography and the characteristics of life style. Based on the clinical knowledge, the multi-dimensional characteristics of sub-health are simplified and summarized, and the decision tree model and the association rule model are used. The research on the longitudinal and transverse relation of the sub-health multi-dimensional features can provide evidence and thought for the sub-health state of the clinical intervention: the intervention system of sub-health population should be carried out in the center of the constitution of the Chinese medicine, In the group of biased constitution, it is possible to intervene with the biased constitution, and the biased population should pay attention to the strengthening of good eating habits and the adjustment of the spleen and stomach. In addition, the biased constitution population is more likely to be combined with the sensitive psychological abnormality of the human relationship, and should be paid attention to timely discharge. For peaceful and qualitative people, although there is no bias of body constitution, it is easy to have the characteristics of traditional Chinese medicine syndrome with qi deficiency and yin deficiency, and can be used for intervention. It should pay attention to good eating habits, education and the combination of people. The geographical and gender characteristics of the place. No matter whether the body is biased or not, the blood deficiency, the syndrome of the kidney and the bladde
【学位授予单位】：广州中医药大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R749.99

【相似文献】