心脏自主神经病变诊断评估、相关危险因素分析及数学模型构建研究

发布时间：2018-01-13 21:03

本文关键词：心脏自主神经病变诊断评估、相关危险因素分析及数学模型构建研究　出处：《复旦大学》2014年博士论文　论文类型：学位论文

【摘要】：研究背景心脏自主神经病变通常被认为是糖尿病比较常见的慢性并发症之一,其发病机制尚不十分明确。现已证实,老年人、冠心病患者、高血压病患者及全身免疫系统疾病患者亦常并发现心脏自主神经病变。心脏自主神经病变具有较高的患病率和发病率。心脏自主神经病变诊断没有真正的金标准存在。Ewing's试验为半定量试验,被用于评价心脏自主神经功能的方法。心率变异性检测是通过记录一定长度的心电图,计算正常窦性心率相邻心搏之间R-R间期的差异程度来反映心率的变化,定量地反映自主神经功能。心率变异性分析方法具有量化、简单可行、客观、安全等优点。无金标准试验下Bayesian方法可以评估诊断实验的性能(如敏感性和特异性)。然而,目前在中国人群中无金标准试验下Bayesian方法评估基于心率变异性标准值心脏自主神经病变诊断性能研究未见报道。人类复杂疾病的发病模型可认为是多基因疾病机制,根据其复杂疾病模型,心脏自主神经病变应该由环境因素、遗传因素及遗传与环境因素交互作用决定。遗传流行病学的大量证据表明心脏自主神经病变是遗传因素和环境因素共同导致的结果。全基因组关联分析方法发现并鉴定了大量与心脏自主神经病变关联的遗传变异。但是中国人群的相关研究鲜有报道,更未见基因-环境交互作用的分析报道。数学模型是使用数学语言描述和抽象事物,是关于部分现实世界和为一种特殊目的而作的一个抽象的、简化的结构。数学模型在临床中的应用通常包括：疾病的筛查模型,风险评估模型及预后模型。疾病定量评估和预测是运用统计方法和数学模型,通过对过去一些历史数据的统计分析,对事物未来的发展趋势、增减速度以及可能达到的发展水平做出数量的说明,并且以数学模型来表达基本规律,对目前状况的评估及对未来发展进行预测。然而,目前没有发现心脏自主神经病变相关的筛查和风险数学模型构建研究。目的本研究目的：1)明确基于短程频率心率变异性标准值的心脏自主神经病变诊断价值评估；2)明确中国人群中心脏自主神经病变相关危险因素(环境、遗传相关因素)；3)构建心脏自主神经病变筛查模型；及4)构建心脏自主神经病变风险模型。方法和结果建立中国人群心脏自主神经病变样本数据库。以多阶段抽样方法(整群抽样及简单随机抽样)征集2092例样本,完成一般数据、生化检查、糖耐量试验、心率变异性检查等临床表型数据收集,提取DNA样本。主要基于横断面研究设计的有关心脏自主神经病变的诊断标准评估,相关危险因素分析,筛查模型及其风险评估模型构建研究。1)无金标准诊断试验下Bayesian分析对心脏自主神经病变诊断评估研究：该研究基于社区的大型横断面数据,研究人群包括2,092受试者,所有受试者完成相应的基线数据收集和短程频率心率变异性测试。同时我们从另一个人群中征集88名同时接受短程频率心率变异性测试和Ewing's试验的受试者。首先从2092个样本人群中选取所有的健康人(371名)明确短程频率心率变异性各组分的参考值。在无诊断金标准情况下,应用Bayesian方法在两个样本中估计基于短程频率心率变异性检测心脏自主神经病变诊断性能。结果显示短程频率心率变异性诊断模型具有很高灵敏度(80%)和特异性(80%)。非劣效性检测表明,短程频率心率变异性的诊断价值并不逊色于Ewing's试验。普通人群中心脏自主神经病变患病率估计为14.92%。在糖尿病患者中,其患病率估计为29.17%。2)中国人群心脏自主神经病变危险因素分析：在心脏自主神经病变样本数据库中,基于诊断评估分析结果,分类患病人群何非患病人群；以Logistic回归模型对收集的临床表型数据进行多因素相关分析,筛选心脏自主神经病变环境相关危险因素；将心脏自主神经病变样本DNA进行SNP分型,获取遗传相关数据,进行基因关联分析；在心脏自主神经病变样本数据库中抽取心脏自主神经病变患者数据进行基因-环境交互作用分析。单因素分析显示,14个风险因素与心脏自主神经病变显著关联(P0.05)。多因素Logistic回归分析5个独立的危险因素：年龄(OR=1.47,95%C1:1.22-1.69, P0.001,表3),心率(OR=2.41,95%CI：2.04-2.71, P 0.001),高血压病病程(O R=1.24,95%CI:1.08-1.41, P0.05),胰岛素抵抗指数(OR=3.45,95%CI:2.12-5.82, P 0.001)和腰围(OR=3.60,95%CI：1.12-6.25,P0.001)。而在本样本中,基因-表型分析表明所选取5个候选基因与心脏自主神经病变无明显关联性,而基因-环境因素交互作用分析表明肥胖表型体重指数与SANIOA (rs7375036)存在交互作用(ORGEI= 5.404,95%CI:1.355-21.558,P=0.017)；糖尿病与SANIOA (rs7375036)存在交互作用(ORGEI=3.453,95%CI:0.973-12.254,P=0.055)；及代谢综合征与ESRI (rs9340799)存在交互作用(ORGEI=1.505,95%CI：0.98-2.312,P=0.062)。3)中国人群心脏自主神经病变筛查模型构建：总样本(2092例样本)被平分为模型生成集和验证集。筛查模型在模型生成集中,由逐步多元Logistic回归分析生成。最终回归模型的变量为心脏自主神经病变筛查模型的组成部分。筛查模型的性能在验证集和总样本中进行评估。最终的筛查模型变量包括年龄、体重指数、高血压病和心率,这些变量与心脏自主神经病变存在显著相关性(P0.05)。模型性能分析表明,在模型的生成集和验证集中,其系统的ROC曲线下面积分别为0.726(95%CI为0.686-0.766)和0.784(95%CI为0.749-0.818)。在验证集中,最佳临界分数为6(风险积分分数范围是0-15),该风险评分系统的敏感性,特异性和需要后续标准测试的比例分别为74.63%、67.50%和39.88%。4)心脏自主神经病变风险模型构建及对比研究：研究目的是应用人工神经网络和多因素Logistic回归模型在自然人群中构建心脏自主神经病变的风险模型,并用比较这两种方法所构建的风险模型的相关性能。将研究样本分为模型生成集和验证集。在同一模型生成集中分别应用神经网络和L ogistic回归模型分析构建相应的心脏自主神经病变风险模型,并在相同的验证集进行评估和预测性能分析。最后将这两类风险模型的性能进行比较。单因素分析显示,14个风险因素与心脏自主神经病变显著关联(P0.05)。Logistic回归构建风险模型的ROC曲线下面积为0.758(95%CI为0.724-0.793),神经网络风险模型ROC曲线下面积为0.762(95%CI为0.732-0.793。结论：研究表明1)短程频率心率变异性参考值可应用于心脏自主神经病变诊断测试,并具有较高灵敏度和特异性。短程频率心率变异性测试对心脏自主神经病变的诊断价值不劣于传统的Ewing's试验,可以应用于心脏自主神经病变的诊断,特别适合于大规模人群诊断应用。心脏自主神经病变在中国人群中具有较高患病率,并在糖尿病、高血压和代谢综合征患者中的患病率更高。2)多因素Logistic回归分析表明心脏自主神经病变的环境相关危险因素表明年龄、心率、高血压病病程及代谢性因素(腰围和胰岛素抵抗指数)与该疾病相关。在本样本中,基因-表型分析表明候选基因与心脏自主神经病变无明显的相关性,但是基因-环境交互分析表明SCNIOA和ESR1基因与代谢性因素存在交互作用。3)我们开发了基于一组简单变量(不需要实验室检查或复杂的临床检查)的心脏自主神经病变的筛查模型。该模型是一种简单、快速、便宜、非侵入性的、可靠的筛查工具。在中国人群中,可应用于该疾病的早期预防,用以延缓疾病的进展。4)本研究有效的应用人工神经网络和Logistic回归构建具有更高区分度和精准度的心脏自主神经病变风险模型,非劣效性检测发现人工神经网络预测模型的灵敏度,特异性和预测值不劣于Logistic回归构建的风险模型。表明这两类风险模型都是有效的评估和预测工具。
[Abstract]:The research background of cardiac autonomic neuropathy is often considered one of the more common chronic complications of diabetes, its pathogenesis is still unclear. It has been confirmed that elderly patients with coronary heart disease, and patients with hypertension and systemic diseases of the immune system is often found and cardiac autonomic neuropathy. Cardiac autonomic neuropathy has a high prevalence and incidence the rate of diagnosis of cardiac autonomic neuropathy. There is no real gold standard.Ewing's test for semi quantitative test method to be used for evaluation of cardiac autonomic function. HRV detection is by ECG recording length and calculate the degree of difference between normal sinus heart rate adjacent cardiac R-R interval to reflect the changes in heart rate, reflect the autonomic nerve function. The heart rate variability analysis method is simple and feasible, quantitative, objective, security and other advantages. No gold standard test Bayesian method can evaluate the performance of diagnostic tests (such as sensitivity and specificity). However, at present in the China crowd no gold standard test method of Bayesian assessment of heart rate variability in the standard value of cardiac autonomic neuropathy diagnosis performance has not been reported. Based on the complex human disease model of disease disease is considered to be a mechanism of polygenic disease, according to the the model of complex diseases, cardiac autonomic neuropathy should be decided by environmental factors, genetic factors and the interaction of genetic and environmental factors. A lot of evidence of Genetic Epidemiology showed that cardiac autonomic neuropathy is a common cause of genetic and environmental factors. The results of a genome-wide association analysis method and found substantial genetic variability and cardiac autonomic neuropathy associated. Identification of Chinese population. But there are few reports on related analysis reports, no more gene environment interaction. The model is the use of mathematical language to describe and abstract things, is on the part of the real world and for a special purpose of an abstract, simplified structure. The application of mathematical model in clinical practice usually includes: screening models of disease, risk assessment model and the prognosis of disease model. The quantitative assessment and prediction is the use of statistics method and mathematical model of the past through some statistical analysis of historical data, the development trend of the things, and increase or decrease the speed may reach the development level of the number of instructions to make, and the expression of basic rules on the mathematical model, the assessment of the current situation and predict the future development. However, there is no research on the construction of screening found the mathematical model and the risk of cardiac autonomic neuropathy related. Purpose: the purpose of this study: 1) based on the standard of God clear cardiac autonomic heart rate variability short frequency value The diagnostic value evaluation; 2) determine the related risk China center crowd dirty autonomic neuropathy (environmental factors, genetic factors); 3) construction of cardiac autonomic neuropathy screening model; and 4) the construction of cardiac autonomic neuropathy risk model. Methods and results to establish China groups of cardiac autonomic neuropathy in multistage sample database. Sampling methods (cluster sampling and simple random sampling) for 2092 samples, the completion of the general data, biochemical examination, glucose tolerance test, heart rate variability examination of clinical phenotype data collection, extraction of DNA samples. The main diagnostic criteria of cross-sectional study about design of cardiac autonomic neuropathy based on the assessment, analysis of the related risk factors. Study on.1 model screening model and risk assessment) no gold standard diagnostic test Bayesian analysis and evaluation study on the diagnosis of cardiac autonomic neuropathy: the study Large cross section data based on community, the study population consisted of 2092 subjects, all subjects completed baseline data collection and short frequency of heart rate variability in the corresponding test. At the same time we from another group of 88 to solicit and accept short frequency heart rate variability test and Ewing's test subjects. The first selection of healthy people all from a sample of 2092 people (371) were clear HRV frequency short-range reference value. In the absence of gold standard in the diagnosis of cases, the application of Bayesian method to estimate the frequency of short-range HRV detection of cardiac autonomic neuropathy diagnosis based on performance in two samples. The results showed that the frequency of short heart rate variability diagnosis model it has a very high sensitivity (80%) and specificity (80%). The results showed noninferiority testing, short frequency of heart rate variability in diagnostic value is inferior to Ewing's in the general population test. Cardiac autonomic neuropathy prevalence estimates of 14.92%. in patients with diabetes, the prevalence rate is estimated at 29.17%.2) analysis of the risk of cardiac autonomic neuropathy factors Chinese crowd: in cardiac autonomic neuropathy sample database, diagnosis and evaluation based on the results of the analysis, classification of population where the prevalence of non Logistic regression; clinical phenotype data collection model multi factor correlation analysis, to screen the environmental risk of cardiac autonomic neuropathy factors; the cardiac autonomic neuropathy samples DNA SNP types, access to genetic data, genetic correlation analysis; data extraction in patients with cardiac autonomic neuropathy in cardiac autonomic neuropathy in the sample database for analysis of gene environment interaction by single factor analysis. Show that the 14 risk factors associated with cardiac autonomic neuropathy (P0.05). Multivariate Logistic Regression analysis of 5 independent risk factors: age (OR=1.47,95%C1:1.22-1.69, P0.001, table 3), heart rate (OR=2.41,95%CI:2.04-2.71, P 0.001), hypertension (O R=1.24,95%CI:1.08-1.41 P0.05), insulin resistance index (OR=3.45,95%CI:2.12-5.82, P 0.001) and waist circumference (OR=3.60,95%CI:1.12-6.25, P0.001). In this sample, genotype phenotype analysis show that the selected 5 candidate genes and cardiac autonomic neuropathy had no obvious relevance, and gene environment interaction analysis showed that BMI and obesity phenotypes of SANIOA (rs7375036) interaction (ORGEI= 5.404,95%CI:1.355-21.558, P=0.017); diabetes mellitus (rs7375036) and SANIOA interaction (ORGEI=3.453,95%CI:0.973-12.254, P=0.055); and the metabolic syndrome ESRI (rs9340799) and interaction (ORGEI= 1.505,95%CI:0.98-2.312, P=0.062).3) China people heart The dirty construction of autonomic neuropathy screening model: the total sample (2092 samples) were divided into model generation and validation set. The screening model generation focused on model analysis. By stepwise Logistic regression to generate part of the final regression model variables for cardiac autonomic neuropathy screening model. The performance of the model is evaluated in the validation screening set and the total sample. The final screening model variables including age, BMI, hypertension disease and heart rate, these variables and cardiac autonomic neuropathy (P0.05). There was significant correlation analysis showed that the performance of the model and focus on generating set and validation of the model, the system under the ROC curve were 0.726 (95%CI to 0.686-0.766) and 0.784 (95%CI 0.749-0.818). In the validation set, the best critical score of 6 (0-15 risk score score range), the sensitivity of risk scoring system, and the specific needs The following standard test respectively 74.63%, 67.50% and 39.88%.4) of cardiac autonomic neuropathy risk model and comparison. The purpose of this study is the application of artificial neural network and multi factor Logistic regression model to construct risk model of cardiac autonomic neuropathy in natural populations, and the related performance risk model constructed by the two methods. The study sample was divided into model generation and validation set. Generation focus respectively using neural network and L ogistic regression model to analyze the construction of cardiac autonomic neuropathy risk model in the same model, and in the same test set to evaluate and predict the performance analysis. Finally compare the performance of the two kinds of risk models the single factor analysis showed that 14 risk factors significantly associated with cardiac autonomic neuropathy (P0.05) ROC curve regression to establish the risk model under.Logistic The area is 0.758 (95%CI 0.724-0.793), ROC curve area of neural network risk model was 0.762 (95%CI to 0.732-0.793. conclusion: 1) short frequency of heart rate variability in the reference value can be used in the diagnosis of cardiac autonomic neuropathy test, and has high sensitivity and specificity. Ewing's test frequency short heart rate variability test diagnostic value the cardiac autonomic neuropathy is not inferior to the traditional, can be used in the diagnosis of cardiac autonomic neuropathy, especially suitable for large populations. Diagnosis of cardiac autonomic neuropathy has a higher prevalence in China populations, and in diabetes, hypertension and metabolic syndrome in patients with a higher prevalence of.2 multi factor Logistic) the regression analysis shows that the environmental risk factors of cardiac autonomic neuropathy showed that age, heart rate, hypertension and metabolic factors (waist circumference and insulin Resistance index) associated with this disease. In this sample, genotype phenotype analysis indicated that the candidate gene and cardiac autonomic neuropathy showed no obvious correlation, but the gene environment interaction analysis showed that SCNIOA and ESR1 genes and metabolic factors interaction.3) we developed based on a simple set of variables (do not need clinical examination laboratory or complex) screening model of cardiac autonomic neuropathy. This model is a simple, fast, inexpensive, non-invasive and reliable screening tool. In the Chinese crowd, can be used in the early stage of the disease prevention, with the progress of.4 to delay disease) this study of the application of artificial neural network and Logistic regression of cardiac autonomic neuropathy risk models more discriminative and accurate, non inferiority detection sensitivity prediction model of artificial neural network, specificity and prediction The value is not inferior to the risk model constructed by Logistic regression. It shows that these two types of risk models are effective evaluation and prediction tools.

【学位授予单位】：复旦大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R541

【参考文献】