缺血性中风病重复测量设计定性数据变化规律的研究
|
摘要
【目的】针对缺血性中风病重复测量设计定性数据，以往的多数研究只停留在横断面分析上，欠缺纵向分析，也没有一套系统而又全面地用于处理此类数据的统计分析方法。本文对此类资料进行深入地研究，试图给出相应数据分析的新思路，挖掘缺血性中风病患者证候随时间变化的规律，揭示中风病的病机本质，帮助并指导临床医生科学地对此病患者实施中医药干预，以期总结出相应的研究方法，为其他疾病获得的重复测量设计定性数据变化规律的研究奠定基础。此外，对中风病证候要素评价量表进行条目筛选、优化量表。本研究旨在为临床研究和实践提供统计学方法的依据和支持。
【内容】本研究主要从证候演变规律、评价量表条目筛选两个方面进行了大规模的统计分析。基于多时点、连续、动态采集的数据，分析缺血性中风病住院患者的内风、内火、痰湿、血瘀、气虚和阴虚6个证候随时间推移的变化规律并寻找对此证候有影响的因素；分别根据患者的首发证候、6个证候各时间点的情况对缺血性中风病患者进行聚类，从而分析各分类中患者的证候随时间的变化规律和相应的影响因素，帮助临床医生找到中药干预的最佳作用点，探讨重复测量设计定性数据变化规律的分析方法。此外，进行中风病证候要素评价量表（含目珠游动、抽搐、头晕、心烦、发热、滑脉等97个症状）条目的筛选研究，探讨项目反应理论在评价量表条目筛选中的应用。
本研究重点针对缺血性中风病重复测量设计定性数据变化规律研究中尚存在的不足进行探讨，借助SAS软件和Mplus软件的编程语言，实现对缺血性中风病患者自身的证候演变规律（证候随时间的变化规律）的挖掘分析以及项目反应理论在中风病证候要素评价量表条目筛选中的应用。
【方法】本研究充分运用多种统计分析方法，特别是广义估计方程、潜在类别分析、潜在转移分析、项目反应理论等。基于北京中医药大学东直门医院所提供的研究基础（国家重点基础研究发展计划课题——缺血性中风病证结合的诊断标准与疗效评价体系研究，课题编号：2003CB517102；国家科技重大专项“重大新药创制”课题——显示中医药疗效优势的中药临床疗效评价关键技术研究，课题编号：2009ZX09502-028），在探索缺血性中风病证候演变模式的研究中，先分别将观测时间作为分类变量和连续变量，单独使用广义估计方程探讨影响本课题收集的全部缺血性中风病患者证候（内风、内火、痰湿、气虚、血瘀和阴虚共6个证候）的因素，以及这6个证候随时间推移的变化规律。之后再根据潜在类别分析和广义估计方程相结合、潜在转移分析和广义估计方程相结合的策略，将观测时间分别作为分类变量和连续变量，分别探索导致不同类别缺血性中风病患者出现不同证候的影响因素以及证候随时间变化的规律。在对中风病证候要素评价量表条目筛选的研究中，使用项目反应理论，通过项目信息函数、区分度参数、项目特征曲线图并综合现实中医理论对评价量表进行条目的筛选，剔除信息量较低的条目，构建logistic曲线回归方程，对项目反应理论最常用的两个参数估计方法（最大似然估计法和贝叶斯估计法）得到的预测概率与真实频率进行比较，根据残差平方和与相关系数找出本研究中最优的参数估计方法。
【结果】本研究对缺血性中风病重复测量设计定性数据研究中现有分析方法存在的不足进行改进，有针对性地提出缺血性中风病证候随时间的变化规律和中风病证候要素评价量表条目筛选的研究策略，通过SAS软件和Mplus软件进行大量编程使分析策略得以实现，通过最适合的形式呈现出来。具体来说，论文的研究结果和主要创新点包括下面四个方面。
（1）对于缺血性中风病重复测量设计定性数据，将受试对象内部相关考虑进来，使用广义估计方程来进行分析。分别将观测时间作为分类变量（重在孤立地看每一个时间点与起点相比对结果的影响，是站在局部的角度来看问题）和连续变量（重在看发生概率随时间的推移的变化规律，是站在全局的角度来看问题），运用广义估计方程对缺血性中风病患者证候数据进行研究，得出影响缺血性中风病患者各证候的因素以及不同证候随时间推移的变化趋势，并可通过拟合的方程对某患者在某时间点出现某证候的概率进行预测。广义估计方程可以较便捷地分析每一个受试对象在不同时间点上被重复观测而得到的缺乏独立性的缺血性中风病重复测量定性数据。
（2）由于发病第一时间出现的证候（首发证候）在临床上有重要意义，对缺血性中风病证候随时间的变化规律的研究摸索出一套分析策略：根据潜在类别分析和广义估计方程相结合的思路，先通过潜在类别分析将缺血性中风病患者按第一时间的6个证候进行聚类，根据拟合指标，聚成2类时最好，“无明显内火”组379例、“内火”组614例。再分别将观测时间作为分类变量和连续变量，运用广义估计方程对这两类缺血性中风病患者证候数据进行研究，得出对首发证候不同的2类患者的6个证候有影响的因素，以及各类患者证候随时间推移的变化规律，并可通过拟合的方程对每组患者在某时间点出现某证候的概率进行预测。根据潜在类别分析聚成的2组患者，各证候的发生率及随时间的变化趋势不尽相同。
（3）全面考虑6个证候各时间点的情况，对缺血性中风病证候随时间的变化规律的研究摸索出一套分析策略：根据潜在转移分析和广义估计方程相结合的思路，先通过潜在转移分析将缺血性中风病患者按6个证候各时间点的情况进行聚类，根据拟合指标，聚成7类时最好，第1类“内风+内火+痰湿+血瘀”组498例、第2类“血瘀+内风”组251例、第3类“阴虚”组87例、第4类“血瘀+内火+痰湿”组63例、第5类“气虚”组52例、第6类“内火+痰湿”组26例、第7类“血瘀”组16例。再分别将观测时间作为分类变量和连续变量，运用广义估计方程对所占比重最大的两类缺血性中风病患者证候数据进行研究，得出对各类患者的6个证候有影响的因素、各类患者证候随时间推移的变化规律以及相邻时间点潜在状态的转移概率，并可通过拟合的方程对每组患者在某时间点出现某证候的概率进行预测。最终发现：各组患者各证候的发生率及随时间的变化趋势不尽相同。
（4）对于中风病证候要素评价量表，使用项目反应理论，得出各项目的难度参数、区分度参数、项目信息函数值、各证候的测验信息函数值和患者能力参数估计值，并绘制各证候下各症状的项目特征曲线、各症状的项目信息函数曲线及各分量表测验特征曲线。通过项目信息函数、区分度参数、项目特征曲线图并综合现实中医理论对评价量表进行条目的筛选，剔除信息量较低f6（项强）、f13（舌短缩）、h24（疾脉）、h25（滑脉）、t10（恶心呕吐）、q18（细脉）、y11（弦脉）和y12（细脉）共8个条目，占总条目数的8.25%。根据条目的参数构建logistic曲线回归方程，可代入患者的能力参数估计值，得到患者各条目水平的概率。对项目反应理论最常用的两种参数估计方法（最大似然估计法和贝叶斯估计法）得到的预测概率与真实频率进行比较，根据残差平方和与相关系数可以看出，本研究中最大似然估计法得到的结果与贝叶斯估计法基本一致，略优于贝叶斯法。
【结论】本研究对缺血性中风病重复测量设计定性数据进行了探索性分析，得到了令人满意的结果。解决了结果变量为证候数据的多元定性资料以及结果变量为证候数据的重复测量设计多元定性资料的受试对象的聚类问题，考虑了受试对象的内部相关性，统计推论更可靠，为研究其他疾病而获得的重复测量设计定性数据的变化规律奠定了基础。在对受试对象聚类后研究每一类患者各证候随时间的变化规律，这对临床医生更有针对性地因病施治，继而提高疗效有重要意义。此外，项目反应理论现在主要应用在心理测量学领域，本研究运用此方法对中风病证候要素评价量表条目进行筛选，结果证明可行，拓宽了项目反应理论的应用领域。
【Objective】As for repeated-measured qualitative data on ischemia stroke, mostprevious studies focused on the cross-sectional analysis. However, the longitudinalanalysis for this kind of data is lacking, so are systematic and comprehensivestatistical methods. This paper, is intended to explore the syndrome variationregularity of ischemia stroke to provide new tactics for data analysis by researchingthe repeated-measured qualitative data on ischemia stroke, which can help unveil themechanism of ischemia stroke and guide clinical intervention. This paper also aims tosummarize relevant analytical methods for studying the variation regularity ofrepeated-measured qualitative data on other diseases. In addition, this paper conductsitem selection and optimization of the stroke syndrome factor evaluation scale. Theoverall purpose of this study is to provide a basis and support for statistical methodsfor clinical research and practice.
【Content】This study performs a large-scale statistical analysis in two aspects:the syndrome variation regularity and item selection of the evaluation scale. To bespecific, this paper explores the variation regularity of six different syndromes ofischemia stroke (including the wind syndrome, fire syndrome, phlegm syndrome,blood stasis syndrome, qi-deficiency syndrome and yin-deficiency syndrome) andseeks influential factors based on continuous and dynamic data. The patients areclassified by the first syndromes and6syndromes at each time point respectively inorder to explore the syndrome variation regularity and the influential factors ofischemia stroke in different classes, help clinicians find the best time for medicalintervention, and explore analytical methods for variation regularity of repeated-measured qualitative data. In addition, this paper conducts item selection of the strokesyndrome factor evaluation scale which contains97symptoms, including dizziness,upset, fever, and discusses the application of item response theory in item selection ofthe evaluation scale.
This study, focused on the drawbacks of variation regularity research onrepeated-measured qualitative data of ischemia stroke, conducts analysis of variationregularity of repeated-measured qualitative data on ischemia stroke, and appliesitem response theory to the item selection of stroke syndrome factor evaluation scalethrough programming language of SAS software and Mplus software.
【Methods】The paper makes full use of various analytical methods of statistics,especially the generalized estimating equation, latent class analysis, latent transition analysis, item response theory. Based on the basic research conducted by DongzhimenHospital Affiliated to Beijing University of Chinese Medicine (“Research on theDiagnostic Criteria and the Efficacy Evaluation System of the symptoms of IschemicStroke”, the project supported by the National Basic Research Program of China,Grant No.:2003CB517102, and “Research on the key techniques in clinical efficacyevaluation that displays the therapeutic advantages of traditional Chinese medicine”,the project supported by the National Science and Technology Major Project,“Significant New Drug Creation”, Grant No.:2009ZX09502-028), we have come toregard the observation time as a classification variable and a continuous variablerespectively in the research on the syndrome variation regularity of ischemia stroke,and adopt the GEE to explore the influential factors and syndrome variation regularityof993ischemia stroke patients. Afterwards, according to the tactic of combiningLCA with GEE, and combining LTA with GEE, we consider the observation time aclassification variable and a continuous variable respectively to explore both theinfluential factors which cause different syndromes of ischemia stroke patients indifferent classes and the syndrome variation regularity. In the study of item selectionof stroke syndrome factor evaluation scale, the item response theory is adopted. Theitem information function, discrimination parameter, item characteristic curve andpractical theory of TCM are used to select items, eliminate items that are relativelyless informative, construct a logistic curve regression equation, compare the predictedprobabilities gained by adopting the two main parameter estimation methods of IRT(the maximum likelihood estimation method and the Bayes estimation method) withthe real frequency, and find the best parameter estimation method according to theresidual sum of squares and the correlation coefficient.
【Results】The paper attempts to eliminate the drawbacks in the existinganalytical methods of repeated-measured qualitative data on ischemia stroke, proposesome tactics for research on syndrome variation regularity of ischemia stroke and itemselection of stroke syndrome factor evaluation scale, and present them in the mostsuitable way through programming by SAS and Mplus. To be specific, the results andmajor innovations of this paper are summarized as follows.
(1) The inner correlation of subjects is taken into consideration, andrepeated-measured qualitative data on ischemia stroke are analyzed using GEE. Theobservation time is considered a classification variable which focuses on the influenceof each time point compared with the starting point (standing in the local point ofview) and a continuous variable which considers the variation regularity of occurrenceprobability with time (standing in the global point of view) respectively. GEE isadopted to explore the influential factors, discuss the syndrome variation regularity ofischemia stroke, and predict the chance that a patient may develop a specificsyndrome at each time point through a fitting equation. GEE can be adopted to analyze repeated-measured qualitative data on ischemia stroke which lacksindependence.
(2) An analytical strategy for syndrome variation regularity of ischemia stroke ispresented based on the fact that the first syndromes are of great clinical significance.By combining LCA with GEE, we classify the patients based on the first6syndromesby LCA. According to the fitting indexes, patients are preferably classified into twogroups, with379and614patients in each group respectively. Afterwards, we considerthe observation time a classification variable and a continuous variable respectively,explore the influential factors and the syndrome variation regularity of ischemiastroke in different classes using GEE, and predict the prospects that a patient in eachclass may develop a certain syndrome at each time point through a fitting equation.The result shows that the syndrome incidence and the variation regularity of the twogroups are different.
(3) An analytical strategy for syndrome variation regularity of ischemia stroke ispresented based on the6syndromes at each time point. By combining LTA with GEE,we classify the patients by the6syndromes at each time point using LTA. The fittingindexes show that it is the best when seven classes are classified, with498,251,87,63,52,26and16patients in each class respectively. This paper regards theobservation time as a classification variable and a continuous variable respectively,afterwards, explore the influential factors, the syndrome variation regularity and thetransition probability by analyzing the two classes that account for the largestproportions using GEE, and predict the probability that a patient in each class maydevelop a particular syndrome at each time point through a fitting equation. The resultsuggests that the syndrome incidence and the variation regularity in each class vary.
(4) The IRT is adopted to acquire difficulty parameter, discrimination parameter,information function, test scores of each syndrome and ability parameter estimates ofpatients, and to draw the item characteristic curve and the test characteristic curve forstroke syndrome factor evaluation scale. Items are selected by item informationfunction, discrimination parameter, item characteristic curve and practical theory ofTCM. As a result, eight items that provide little information (including f6, f13, h24,h25, t10, q18, y11and y12) are eliminated, which account for8.25%of the total. Alogistic curve regression equation is constructed using item parameters so that thechance of each patient having each item can be obtained by inputting the abilityparameter estimate of each patient. Afterwards, the predicted probabilities acquired byadopting the two most-frequently-used parameter estimation methods of IRT (themaximum likelihood estimation method and the Bayes estimation method) arecompared with the real frequency. According to the residual sum of squares and thecorrelation coefficient, we are able to draw the conclusion that the results of the abovetwo methods are consistent, and the MLE method is marginally better than the Bayes method.
【Conclusions】The paper explores repeated-measured qualitative data ofischemia stroke and achieves some desirable results by resolving the issue of sampleclassification for qualitative data and repeatedly-measured quantitative data withmultiple response variables that are syndromes. The inner correlation of subjects istaken into consideration in this study so that the statistical inference is highly reliable.What’s more, it lays the foundation for studying the variation regularity ofrepeated-measured qualitative data of other diseases. This paper discusses thevariation regularity of each class after sample clustering, which helps guide clinicalintervention at different stages for patients suffering from ischemia stroke in differentclasses, and improve the curative effect. In addition, this paper uses the item responsetheory, which is mainly used in the field of psychological measurement presently, initem selection of stroke syndrome factor evaluation scale. It is proved that the result isfeasible, which indicates that the application of IRT is extended.
【内容】本研究主要从证候演变规律、评价量表条目筛选两个方面进行了大规模的统计分析。基于多时点、连续、动态采集的数据，分析缺血性中风病住院患者的内风、内火、痰湿、血瘀、气虚和阴虚6个证候随时间推移的变化规律并寻找对此证候有影响的因素；分别根据患者的首发证候、6个证候各时间点的情况对缺血性中风病患者进行聚类，从而分析各分类中患者的证候随时间的变化规律和相应的影响因素，帮助临床医生找到中药干预的最佳作用点，探讨重复测量设计定性数据变化规律的分析方法。此外，进行中风病证候要素评价量表（含目珠游动、抽搐、头晕、心烦、发热、滑脉等97个症状）条目的筛选研究，探讨项目反应理论在评价量表条目筛选中的应用。
本研究重点针对缺血性中风病重复测量设计定性数据变化规律研究中尚存在的不足进行探讨，借助SAS软件和Mplus软件的编程语言，实现对缺血性中风病患者自身的证候演变规律（证候随时间的变化规律）的挖掘分析以及项目反应理论在中风病证候要素评价量表条目筛选中的应用。
【方法】本研究充分运用多种统计分析方法，特别是广义估计方程、潜在类别分析、潜在转移分析、项目反应理论等。基于北京中医药大学东直门医院所提供的研究基础（国家重点基础研究发展计划课题——缺血性中风病证结合的诊断标准与疗效评价体系研究，课题编号：2003CB517102；国家科技重大专项“重大新药创制”课题——显示中医药疗效优势的中药临床疗效评价关键技术研究，课题编号：2009ZX09502-028），在探索缺血性中风病证候演变模式的研究中，先分别将观测时间作为分类变量和连续变量，单独使用广义估计方程探讨影响本课题收集的全部缺血性中风病患者证候（内风、内火、痰湿、气虚、血瘀和阴虚共6个证候）的因素，以及这6个证候随时间推移的变化规律。之后再根据潜在类别分析和广义估计方程相结合、潜在转移分析和广义估计方程相结合的策略，将观测时间分别作为分类变量和连续变量，分别探索导致不同类别缺血性中风病患者出现不同证候的影响因素以及证候随时间变化的规律。在对中风病证候要素评价量表条目筛选的研究中，使用项目反应理论，通过项目信息函数、区分度参数、项目特征曲线图并综合现实中医理论对评价量表进行条目的筛选，剔除信息量较低的条目，构建logistic曲线回归方程，对项目反应理论最常用的两个参数估计方法（最大似然估计法和贝叶斯估计法）得到的预测概率与真实频率进行比较，根据残差平方和与相关系数找出本研究中最优的参数估计方法。
【结果】本研究对缺血性中风病重复测量设计定性数据研究中现有分析方法存在的不足进行改进，有针对性地提出缺血性中风病证候随时间的变化规律和中风病证候要素评价量表条目筛选的研究策略，通过SAS软件和Mplus软件进行大量编程使分析策略得以实现，通过最适合的形式呈现出来。具体来说，论文的研究结果和主要创新点包括下面四个方面。
（1）对于缺血性中风病重复测量设计定性数据，将受试对象内部相关考虑进来，使用广义估计方程来进行分析。分别将观测时间作为分类变量（重在孤立地看每一个时间点与起点相比对结果的影响，是站在局部的角度来看问题）和连续变量（重在看发生概率随时间的推移的变化规律，是站在全局的角度来看问题），运用广义估计方程对缺血性中风病患者证候数据进行研究，得出影响缺血性中风病患者各证候的因素以及不同证候随时间推移的变化趋势，并可通过拟合的方程对某患者在某时间点出现某证候的概率进行预测。广义估计方程可以较便捷地分析每一个受试对象在不同时间点上被重复观测而得到的缺乏独立性的缺血性中风病重复测量定性数据。
（2）由于发病第一时间出现的证候（首发证候）在临床上有重要意义，对缺血性中风病证候随时间的变化规律的研究摸索出一套分析策略：根据潜在类别分析和广义估计方程相结合的思路，先通过潜在类别分析将缺血性中风病患者按第一时间的6个证候进行聚类，根据拟合指标，聚成2类时最好，“无明显内火”组379例、“内火”组614例。再分别将观测时间作为分类变量和连续变量，运用广义估计方程对这两类缺血性中风病患者证候数据进行研究，得出对首发证候不同的2类患者的6个证候有影响的因素，以及各类患者证候随时间推移的变化规律，并可通过拟合的方程对每组患者在某时间点出现某证候的概率进行预测。根据潜在类别分析聚成的2组患者，各证候的发生率及随时间的变化趋势不尽相同。
（3）全面考虑6个证候各时间点的情况，对缺血性中风病证候随时间的变化规律的研究摸索出一套分析策略：根据潜在转移分析和广义估计方程相结合的思路，先通过潜在转移分析将缺血性中风病患者按6个证候各时间点的情况进行聚类，根据拟合指标，聚成7类时最好，第1类“内风+内火+痰湿+血瘀”组498例、第2类“血瘀+内风”组251例、第3类“阴虚”组87例、第4类“血瘀+内火+痰湿”组63例、第5类“气虚”组52例、第6类“内火+痰湿”组26例、第7类“血瘀”组16例。再分别将观测时间作为分类变量和连续变量，运用广义估计方程对所占比重最大的两类缺血性中风病患者证候数据进行研究，得出对各类患者的6个证候有影响的因素、各类患者证候随时间推移的变化规律以及相邻时间点潜在状态的转移概率，并可通过拟合的方程对每组患者在某时间点出现某证候的概率进行预测。最终发现：各组患者各证候的发生率及随时间的变化趋势不尽相同。
（4）对于中风病证候要素评价量表，使用项目反应理论，得出各项目的难度参数、区分度参数、项目信息函数值、各证候的测验信息函数值和患者能力参数估计值，并绘制各证候下各症状的项目特征曲线、各症状的项目信息函数曲线及各分量表测验特征曲线。通过项目信息函数、区分度参数、项目特征曲线图并综合现实中医理论对评价量表进行条目的筛选，剔除信息量较低f6（项强）、f13（舌短缩）、h24（疾脉）、h25（滑脉）、t10（恶心呕吐）、q18（细脉）、y11（弦脉）和y12（细脉）共8个条目，占总条目数的8.25%。根据条目的参数构建logistic曲线回归方程，可代入患者的能力参数估计值，得到患者各条目水平的概率。对项目反应理论最常用的两种参数估计方法（最大似然估计法和贝叶斯估计法）得到的预测概率与真实频率进行比较，根据残差平方和与相关系数可以看出，本研究中最大似然估计法得到的结果与贝叶斯估计法基本一致，略优于贝叶斯法。
【结论】本研究对缺血性中风病重复测量设计定性数据进行了探索性分析，得到了令人满意的结果。解决了结果变量为证候数据的多元定性资料以及结果变量为证候数据的重复测量设计多元定性资料的受试对象的聚类问题，考虑了受试对象的内部相关性，统计推论更可靠，为研究其他疾病而获得的重复测量设计定性数据的变化规律奠定了基础。在对受试对象聚类后研究每一类患者各证候随时间的变化规律，这对临床医生更有针对性地因病施治，继而提高疗效有重要意义。此外，项目反应理论现在主要应用在心理测量学领域，本研究运用此方法对中风病证候要素评价量表条目进行筛选，结果证明可行，拓宽了项目反应理论的应用领域。
【Objective】As for repeated-measured qualitative data on ischemia stroke, mostprevious studies focused on the cross-sectional analysis. However, the longitudinalanalysis for this kind of data is lacking, so are systematic and comprehensivestatistical methods. This paper, is intended to explore the syndrome variationregularity of ischemia stroke to provide new tactics for data analysis by researchingthe repeated-measured qualitative data on ischemia stroke, which can help unveil themechanism of ischemia stroke and guide clinical intervention. This paper also aims tosummarize relevant analytical methods for studying the variation regularity ofrepeated-measured qualitative data on other diseases. In addition, this paper conductsitem selection and optimization of the stroke syndrome factor evaluation scale. Theoverall purpose of this study is to provide a basis and support for statistical methodsfor clinical research and practice.
【Content】This study performs a large-scale statistical analysis in two aspects:the syndrome variation regularity and item selection of the evaluation scale. To bespecific, this paper explores the variation regularity of six different syndromes ofischemia stroke (including the wind syndrome, fire syndrome, phlegm syndrome,blood stasis syndrome, qi-deficiency syndrome and yin-deficiency syndrome) andseeks influential factors based on continuous and dynamic data. The patients areclassified by the first syndromes and6syndromes at each time point respectively inorder to explore the syndrome variation regularity and the influential factors ofischemia stroke in different classes, help clinicians find the best time for medicalintervention, and explore analytical methods for variation regularity of repeated-measured qualitative data. In addition, this paper conducts item selection of the strokesyndrome factor evaluation scale which contains97symptoms, including dizziness,upset, fever, and discusses the application of item response theory in item selection ofthe evaluation scale.
This study, focused on the drawbacks of variation regularity research onrepeated-measured qualitative data of ischemia stroke, conducts analysis of variationregularity of repeated-measured qualitative data on ischemia stroke, and appliesitem response theory to the item selection of stroke syndrome factor evaluation scalethrough programming language of SAS software and Mplus software.
【Methods】The paper makes full use of various analytical methods of statistics,especially the generalized estimating equation, latent class analysis, latent transition analysis, item response theory. Based on the basic research conducted by DongzhimenHospital Affiliated to Beijing University of Chinese Medicine (“Research on theDiagnostic Criteria and the Efficacy Evaluation System of the symptoms of IschemicStroke”, the project supported by the National Basic Research Program of China,Grant No.:2003CB517102, and “Research on the key techniques in clinical efficacyevaluation that displays the therapeutic advantages of traditional Chinese medicine”,the project supported by the National Science and Technology Major Project,“Significant New Drug Creation”, Grant No.:2009ZX09502-028), we have come toregard the observation time as a classification variable and a continuous variablerespectively in the research on the syndrome variation regularity of ischemia stroke,and adopt the GEE to explore the influential factors and syndrome variation regularityof993ischemia stroke patients. Afterwards, according to the tactic of combiningLCA with GEE, and combining LTA with GEE, we consider the observation time aclassification variable and a continuous variable respectively to explore both theinfluential factors which cause different syndromes of ischemia stroke patients indifferent classes and the syndrome variation regularity. In the study of item selectionof stroke syndrome factor evaluation scale, the item response theory is adopted. Theitem information function, discrimination parameter, item characteristic curve andpractical theory of TCM are used to select items, eliminate items that are relativelyless informative, construct a logistic curve regression equation, compare the predictedprobabilities gained by adopting the two main parameter estimation methods of IRT(the maximum likelihood estimation method and the Bayes estimation method) withthe real frequency, and find the best parameter estimation method according to theresidual sum of squares and the correlation coefficient.
【Results】The paper attempts to eliminate the drawbacks in the existinganalytical methods of repeated-measured qualitative data on ischemia stroke, proposesome tactics for research on syndrome variation regularity of ischemia stroke and itemselection of stroke syndrome factor evaluation scale, and present them in the mostsuitable way through programming by SAS and Mplus. To be specific, the results andmajor innovations of this paper are summarized as follows.
(1) The inner correlation of subjects is taken into consideration, andrepeated-measured qualitative data on ischemia stroke are analyzed using GEE. Theobservation time is considered a classification variable which focuses on the influenceof each time point compared with the starting point (standing in the local point ofview) and a continuous variable which considers the variation regularity of occurrenceprobability with time (standing in the global point of view) respectively. GEE isadopted to explore the influential factors, discuss the syndrome variation regularity ofischemia stroke, and predict the chance that a patient may develop a specificsyndrome at each time point through a fitting equation. GEE can be adopted to analyze repeated-measured qualitative data on ischemia stroke which lacksindependence.
(2) An analytical strategy for syndrome variation regularity of ischemia stroke ispresented based on the fact that the first syndromes are of great clinical significance.By combining LCA with GEE, we classify the patients based on the first6syndromesby LCA. According to the fitting indexes, patients are preferably classified into twogroups, with379and614patients in each group respectively. Afterwards, we considerthe observation time a classification variable and a continuous variable respectively,explore the influential factors and the syndrome variation regularity of ischemiastroke in different classes using GEE, and predict the prospects that a patient in eachclass may develop a certain syndrome at each time point through a fitting equation.The result shows that the syndrome incidence and the variation regularity of the twogroups are different.
(3) An analytical strategy for syndrome variation regularity of ischemia stroke ispresented based on the6syndromes at each time point. By combining LTA with GEE,we classify the patients by the6syndromes at each time point using LTA. The fittingindexes show that it is the best when seven classes are classified, with498,251,87,63,52,26and16patients in each class respectively. This paper regards theobservation time as a classification variable and a continuous variable respectively,afterwards, explore the influential factors, the syndrome variation regularity and thetransition probability by analyzing the two classes that account for the largestproportions using GEE, and predict the probability that a patient in each class maydevelop a particular syndrome at each time point through a fitting equation. The resultsuggests that the syndrome incidence and the variation regularity in each class vary.
(4) The IRT is adopted to acquire difficulty parameter, discrimination parameter,information function, test scores of each syndrome and ability parameter estimates ofpatients, and to draw the item characteristic curve and the test characteristic curve forstroke syndrome factor evaluation scale. Items are selected by item informationfunction, discrimination parameter, item characteristic curve and practical theory ofTCM. As a result, eight items that provide little information (including f6, f13, h24,h25, t10, q18, y11and y12) are eliminated, which account for8.25%of the total. Alogistic curve regression equation is constructed using item parameters so that thechance of each patient having each item can be obtained by inputting the abilityparameter estimate of each patient. Afterwards, the predicted probabilities acquired byadopting the two most-frequently-used parameter estimation methods of IRT (themaximum likelihood estimation method and the Bayes estimation method) arecompared with the real frequency. According to the residual sum of squares and thecorrelation coefficient, we are able to draw the conclusion that the results of the abovetwo methods are consistent, and the MLE method is marginally better than the Bayes method.
【Conclusions】The paper explores repeated-measured qualitative data ofischemia stroke and achieves some desirable results by resolving the issue of sampleclassification for qualitative data and repeatedly-measured quantitative data withmultiple response variables that are syndromes. The inner correlation of subjects istaken into consideration in this study so that the statistical inference is highly reliable.What’s more, it lays the foundation for studying the variation regularity ofrepeated-measured qualitative data of other diseases. This paper discusses thevariation regularity of each class after sample clustering, which helps guide clinicalintervention at different stages for patients suffering from ischemia stroke in differentclasses, and improve the curative effect. In addition, this paper uses the item responsetheory, which is mainly used in the field of psychological measurement presently, initem selection of stroke syndrome factor evaluation scale. It is proved that the result isfeasible, which indicates that the application of IRT is extended.
引文
[1]王杰.中医中药治疗缺血性中风研究概况.中医临床研究,2011,3(14):116-117.
[2]刘新胜,崔书克.缺血性中风的中医药研究进展.河南中医,2012,32(1):129-131.
[3]韦桂梅.缺血性中风中医辨证研究近况.云南中医中药杂志,2006,27(3):63-64.
[4]赵国强.缺血性中风药物治疗现状.医学综述,1999,5(4):153-156.
[5]丁自娟.《内经》论治中风.中国中医急症,2010,19(1):111-112.
[6]张填,马涤辉,孙龙.脑卒中患者预后影响因素的分析.医学综述,2007,13(5):385-387.
[7]耿晓娟,张军平,高颖,等.缺血性中风病急性期证候变化规律研究.中华中医药杂志,2010,25(9):1485-1487.
[8]钟海珍,马斌,高颖.缺血性中风病患者发病3个月致残因素初探.中国中医基础医学杂志,2010,16(1):47-50.
[9]马斌,高颖.中风病发病第7天和第14天证候要素演变规律初步研究.辽宁中医杂志,2006,33(12):1561-1563.
[10]王建华.出血性中风、缺血性中风急性期证候演变规律的研究.中国中医急症,2001,10(4):215-217.
[11]郭蕾,王永炎,张志斌.证候概念的诠释.北京中医药大学学报,2003,26(2):5-8.
[12]王永炎,张志斌.再议完善辨证方法体系的几个问题.天津中医药,2007,24(1):1-4.
[13]张志斌,王永炎,吕爱平,等.论证候要素与证候靶点应证组合辨证.中医杂志,2006,47(7):483-485.
[14]刘伍立,欧阳建军,黄博辉.中医文献对中风病的阐述与述评.针灸临床杂志,2006,22(10):5-8.
[15]辛喜燕,张华,高颖.缺血性中风急性期痰热证的证候演变特点及其与神经功能缺损程度的关系.辽宁中医杂志,2010,37(9):1644-1646.
[16]鲍远程,杨文明,胡翠平,等.缺血性中风的中医证候特点研究.中医药临床杂志,2012,24(3):189-192.
[17]杨利,黄燕,蔡业峰,等.1418例中风患者痰瘀证候分布和演变规律探析.辽宁中医杂志,2004,31(6):459-460.
[18]王顺道,杜梦华,解庆凡,等.中风病急性期证候演变规律的研究.中国中医急症,1995,5(3):121-124.
[19]徐俊峰,李保洋,高颖,等.缺血性中风病证候演变模式的研究.天津中医药,2009,26(6):511-514.
[20]谢颖桢,邹忆怀,马洪明,等.中风病急性期中经络和中脏腑证候要素的动态比较.中国中医基础医学杂志,2006,12(10):746-748.
[21]张华.缺血性中风证候演变与神经功能缺损的相关性研究.北京中医药大学学报,2007,30(4):274-278.
[22]刘强,高颖.中风病不同阶段四诊信息的指标聚类分析.中华中医药杂志,2007,22(3):162-164.
[23] Nelder JA, Wedderburn RWM. Generalized linear models. Journal of theRoyal Statistical Society,SeriesA,1972,135:370–384.
[24] Sutradhar BC. An overview on regression models for discrete longitudinalresponses. Statistical Science,2003,18(3):377-393.
[25] Fitzmaurice G, Davidian M, Verbeke G, et al. Longitudinal data analysis.BocaRaton,FL:Chapman&Hall/CRC Press,2009:1-78.
[26] Liang KY, Zeger ST. Longitudinal data analysis using generalized linearmodels. Biometrika,1986,73(1):13-22.
[27] Zeger ST, Liang KY. Longitudinal data analysis for discrete and continuousoutcomes. Biomeics,1986,42:121-130.
[28]高燕宁,蔡文玮,周纪芗.广义估计方程GEE1与纵向资料的回归分析.数理医药学杂志,1994,7(2):118-123.
[29] SAS Institute Inc. SAS/STAT9.2user’s guide. Cary,NC:SAS Institute Inc.,2008:1982-2070.
[30]方积乾.现代医学统计学.北京:人民卫生出版社,2002:25-61.
[31] Liang KY, Zeger SL, Quaqish B. Multivariate regression analysis forcategorical data. Journal of the Royal Statistical Society,Series B,1992,54:3-40.
[32]夏彦,潘晓平,刘元元,等.广义估计方程在临床试验重复测量资料中的应用.现代预防医学,2005,32(5):444-445.
[33] Horton NJ, Lipsitz SR. Review of software to fit generalized estimatingequation regression models. The American Statistician,1999,53:160-169.
[34]朱玉,王静,何倩.广义估计方程在SPSS统计软件中的实现.中国卫生统计,2011,28(2):199-201.
[35]陈启光.纵向研究中重复测量资料的广义估计方程分析.中国卫生统计,1995,12(1):22-25.
[36] Carey VC, Zeger SL, Diggle P. Modelling multivariate binary data withalternating logistic regressions. Biomerika,1993,80(3):517-526.
[37] Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalizedestimating equation approach. Biometrics,1988,44(4):1049-1060.
[38]李运明,封宗超,夏勋,等.广义估计方程在CT显示方法研究数据分析中的应用.现代生物医学进展,2010,10(21):4173-4176.
[39] Akaike H. Factor analysis and the AIC. Psychometrika,1987,52:317–332.
[40] Raftery AE. Bayesian model selection in social research. Sociological Metho-dology,1995,25:111–163.
[41] Park T. A comparison of the generalized estimating equation approach withthe maximum likelihood approach for repeated measurements. Stat Med,1993,12(18):1723-1732.
[42] Pan W. Goodness-of-fit tests for GEE with correlated binary data.Scandinavian Journal of Statistics,2002,29:101-110.
[43] Pan W. Akaike's information criterion in generalized estimating equations.Biometrics,2001,57(1):120-125.
[44] Williamson JM, Lin HM, Barnhart HX. A classification statistic for GEEcategorical response models. Journal of Data Science,2003,1:149-165.
[45] Lin DY, Wei LJ, Ying Z. Checking the cox model with cumulative sums ofmartingale-based residuals. Biometrika,1993,80:557-572.
[46] Lin DY, Wei LJ, Ying Z. Model-checking techniques based on cumulativeresiduals. Biometrics,2002,58:1-12.
[47]罗天娥,赵晋芳,刘桂芬.GENMOD过程和GLIMMIX过程的比较.中国卫生统计,2010,27(2):196-200.
[48]罗天娥,赵晋芳,刘桂芬.累积残差在广义估计方程模型诊断中的应用.中国卫生统计,2009,26(4):387-390.
[49] Lazarsfeld PF. The logical and mathematical foundations of latent structureanalysis. NJ:Princeton University Press,1950:362-412.
[50] Formann AK, Kohlmann T. Latent class analysis in medical research. StatMethods Med Res,1996,5:179–211.
[51]张洁婷,焦璨,张敏强,等.潜在类别分析技术在心理学研究中的应用.心理科学进展,2010,18(12):1991-1998.
[52]申春悌,陈启光,陆岩,等.中医证候要素研究中潜在变量模型的应用.北京中医药大学学报,2010,33(11):725-731.
[53]陈炳为,陈启光,许碧云.潜在变量模型及其在中医证候中的应用概述.中国卫生统计,2009,26(5):535-538.
[54] Collins LM, Lanza ST. Latent class and latent transition analysis: withapplications in the social, behavioral and health sciences. Hoboken,NJ:JohnWiley&Sons,2010:3-224.
[55] Lanza ST, Bary BC. Transitions in drug use among high-risk women: anapplication of latent class and latent transition analysis. Adv Appl Stat Sci,2010,3(2):203-235.
[56] Smith RA, Barclay VC, Smith JLF. Investigating preferences for mosquito-control technologies in mozambique with latent class analysis. Malaria Journal,2011,10:200.
[57] Nylund KL, Asparouhov T, Muthén Bo. Deciding on the number of classes inlatent class analysis and growth mixture modeling: a Monte Carlo simulationstudy. Struct Equ Modeling,2007,14(4):535-569.
[58] Sotres-Alvarez D, Herring AH, Siega-Riz AM. Latent class analysis is usefulto classify pregnant women into dietary patterns. The Journal of Nutrition,2010,140:2253-2259.
[59] Trivedi RB, Ayotte BJ, Thorpe CT, et al. Is there a nonadherent subtype ofhypertensive patient? A latent class analysis approach. Patient Preference andAdherence,2010,4:255-262.
[60] Magidson J, Vermunt JK. Latent class models for clustering: a comparisonwith k-means. CJMR,2002,20:37–44.
[61] Beeber AS, Thorpe JM, Clipp EC. Community-based service use by elderswith dementia and their caregivers: a latent class analysis. Nurs Res,2008,57(5):312–321.
[62] Thorpe JM, Thorpe CT, Kennelty KA, et al. Patterns of perceived barriers tomedical care in older adults: a latent class analysis. BMC Health ServicesResearch,2011,11:181.
[63] Lanza ST, Collins LM, Lemmon DR, et al. PROC LCA: a SAS procedure forlatent class analysis. Structural Equation Modeling,2007,14(4):671-694.
[64] Magidson J, Vermunt JK. Latent class factor and cluster models, bi-plots andtri-plots and related graphical displays. Sociological Mehtodology,2001,31:223-264.
[65]陈炳为,郑水龙,刘沛,等.潜在分类模型在无金标准诊断试验中的评价.中国卫生统计,2010,27(1):11-16.
[66]张岩波.潜变量分析.北京:高等教育出版社,2009:220-246.
[67] Heitzler C, Lytle L, Erickson D, et al. Physical activity and sedentary activitypatterns among children and adolescents: a latent class analysis approach. JPhys Act Health,2011,8(4):457-467.
[68] Boyko EJ, Doheny RA, McNeely MJ, et al. Latent class analysis of themetabolic syndrome. Diabetes Res Clin Pract,2010,89(1):88-93.
[69]郭小玲,裴磊磊,张岩波.潜在类别模型及数据模拟分析.数理医药学杂志,2009,22(6):631-635.
[70] Lanza ST, Dziak JJ, Huang L, et al(2011). PROC LCA&PROC LTA User'sGuide(Version1.2.7). University Park: The Methodology Center, Penn State.Retrieved from http://methodology.psu.edu.
[71]邱皓政.潜在类别模型的原理与技术.北京:教育科学出版社,2008:1-47.
[72] Hagenaars JA, McCutcheon AL. Applied latent class analysis. Cambridge:Cambridge University Press,2002:89-106.
[73] Dean N, Raftery AE. Latent class analysis variable selection. Ann Inst StatMath,2010,62(1):11-35.
[74] Goodman LA. Exploratory latent structure analysis using both identifiable andunidentifiable models. Biometrika,1974,61(2):215-231.
[75] Formann AK. Linear logistic latent class analysis for polytomous data. Journalof the American Statistical Association,1992,87:476-486.
[76] Lin TH, Dayton CM. Model selection information criteria for non-nestedlatent class models. Journal of Education and Behavioral Statistics,1997,22(3):249-264.
[77] Silverwood RJ, Nitsch D, Pierce M, et al. Characterizing longitudinal patternsof physical activity in mid-adulthood using latent class analysis: results from aprospective cohort study. American Journal of Epidemiology,2011,174(12):1406-1415.
[78] Goodman LA. On the assignment of individuals to latent classes. SociologicalMethodology,2007,37(1):1-22.
[79]孟灿,武俊青,李玉艳,等.潜类别分析原理及其在聚类分析中的应用.中国卫生统计,2010,27(3):237-239.
[80] Chung H, Lanza ST, Loken E. Latent transition analysis: inference andestimation. Statistics in Medicine,2008,27(11):1834–1854.
[81] Lanza ST, Patrick ME, Maggs JL. Latent transition analysis: benefits of alatent variable approach to modeling transitions in substance use. J DrugIssues,2010,40(1):93-120.
[82] Lanza ST, Collins LM. A new SAS procedure for latent transition analysis:transitions in dating and sexual risk behavior. Developmental Psychology,2008,44(2):446-456.
[83] Roberts TJ, Ward SE. Using latent transition analysis in nursing research toexplore change over time. Nurs Res,2011,60(1):73-79.
[84] Dempster AP, Laird NM, Rubin DB. Maximum likelihood from incompletedata via EM algorithm(with discussion). Journal of the Royal StatisticalSociety,Series B,1977,39:1–38.
[85] Connell A, Bullock BM, Dishion TJ, et al. Family intervention effects on co-occurring early childhood behavioral and emotional problems: a latenttransition analysis approach. J Abnorm Child Psychol,2008,36(8):1211-1225.
[86] Alagumalai S, Curtis DD, Hungi N. Applied rasch measurement: a book ofexemplars. Netherlands:Springer,2005:1-14.
[87] Novick MR. The axioms and principal results of classical test theory. Journalof Mathematical Psychology,1966,3:1–18.
[88]余民宁.项目反应理论IRT及其应用.台北:心理出版社,2009:1-50.
[89]漆书青,戴海崎.项目反应理论及其应用研究.南昌:江西高校出版社,1992:129-215.
[90] Embretson SE, Reise SP. Item response theory for psychologists. Mahwah,New Jersey:Lawrence Erlbaum Associates,2000:3-246.
[91] Bock RD. Estimating item parameters and latent ability when responses arescored in two or more nominal categories. Psychometrika,1972,37:29–51.
[92] Mellenbergh GJ. Conceptual notes on models for discrete polytomous itemresponses. Applied Psychological Measurement,1995,19:91–100.
[93]高怀勇,金桂林.项目反应理论及其在计算机自适应测试中的应用.西华师范大学学报（自然科学版）,2008,29(1):84-88.
[94] Kosinski M, Bayliss MS, Bjorner JB, et al. A six-item short-form survey formeasuring headache impact: the HIT-6. Quality of Life Research,2003,12(8):963-974.
[95] Green DR, Yen WM, Burket GR. Experience in the application of itemresponse theory in test construction. Applied Measurement in Education,1989,2(4):297-312.
[96] Weiss DJ. Adaptive testing by computer. Journal of Consulting and ClinicalPsychology,1985,53:774-789.
[97] Ware JE, Kosinski M, Bjorner JB, et al. Applications of computerizedadaptive testing (CAT) to the assessment of headache impact. Quality of LifeResearch,2003,12(9):935-952.
[98]查尔斯L赫林.项目反应理论—在心理测量中的应用.武汉:湖北教育出版社,1990:14-82.
[99] Van der Linden WJ, Hambleton RK. Handbook of modern item responsetheory. New York:Springer-Verlag,1997:1-28.
[100]陈炳为,许碧云,陈启光,等.两分类项目反应理论在中医证候中的应用.中国卫生统计,2011,28(1):16-18,21.
[101] Lai JS, Cella D, Chang CH, et al. Item banking to improve, shorten andcomputerize self-reported fatigue: an illustration of steps to create a core itembank from the FACIT-Fatigue Scale. Quality of Life Research,2003,12(5):485-501.
[102] Robitail S, Erhart M, Tebe C, et al. Person fit across European countries:results from the kidscreen field survey. Quality of Life Research,2004,13(9):1547.
[103]许祖慰.项目反应理论及其在测验中的应用.上海:华东师范大学出版社,1992:18-129.
[104] McDonald RP. A basis for multidimensional item response theory. AppliedPsychological Measurement,2000,24:99-114.
[105] Reckase MD. The past and future of multidimensional item response theory.Applied Psychological Measurement,1997,21(1):25-36.
[106] Reckase MD, Ackerman TA, Carlson JE. Building a unidimensional test usingmultidimensional items. Journal of Educational Measurement,1988,25:193-203.
[107]林岳卿,方积乾.多维IRT与单维IRT在多维量表中应用的差异.中国卫生统计,2011,28(3):226-228.
[108] Reeves BB, Hays RD, Bjorner JB, et al. Psychometric evaluation andcalibration of health-related quality of life item banks: plans for the patient-reported outcomes measurement information systems(PROMIS). Med Care,2007,45:22-31.
[109] Davier MV, Carstensen CH. Multivariate and mixture distribution raschmodels: extensions and applications. New York:Springer,2007:1-55.
[110]余嘉元.项目反应理论及其应用.南京:江苏教育出版社,1992:1-157.
[111]杜文久.高等项目反应理论.重庆:西南师范大学出版社,2007:52-277.
[112] Muthén LK, Muthén BO. Mplus user’s guide.6thedition. Los Angeles,CA:Muthén&Muthén,2010:60-61.
[113] Baker FB. The basics of item response theory.2ndedition. College Park,MD:ERIC Clearinghouse on Assessment and Evaluation,2001:5-128.
[114] Baker FB. Methodology reviews: item parameter estimation under the one-,two-, and three-parameter logistic models. Applied Psychological Measure-ment,1987,11:111-142.
[115] Bock RD, Aitkin M. Marginal maximum likelihood estimation of itemparameters: an application of the EM algorithm. Psychometrika,1981,46:443–459.
[116] Kingston NM, Dorans NJ. The analysis of item-ability regression: anexploratory IRT model fit tool. Applied Psychological Measurement,1985,9:281-288.
[117] Ludlow LH. A strategy for the graphical representation of Rasch modelresiduals. Educational and Psychological Measurement,1985,45:851-859.
[118] Yen WM. Using simulation results to choose a latent trait model. AppliedPsychological Measurement,1981,5:245-262.
[119]胡良平.医学统计学—应用三型理论分析定量资料与定性资料.北京:人民军医出版社,2009:325-393.
[120] Cox DR. The analysis of binary data. London:Methuen,1970:14-29.
[121]金丕焕.医用统计方法.第2版.上海:复旦大学出版社,2003:332-343,355-365.
[122]胡良平.医学统计学—运用三型理论进行现代回归分析.北京:人民军医出版社,2010:158-174,203-220.
[123]胡良平.SAS统计分析教程.北京:电子工业出版社,2010:183-191,332-343.
[124]张家放.医用多元统计方法.武汉:华中科技大学出版社,2002:127-157.
[125]方积乾.卫生统计学.第5版.北京:人民卫生出版社,2007:356-367.
[126]陈峰.医用多元统计分析方法.第2版.北京:中国统计出版社,2007:82-113.
[127]柳青.多元统计分册—中国医学统计百科全书.北京:人民卫生出版社,2004:218-221.
[128] Hosmer DW, Lemeshow S. Applied logistic regression.2ndedition.John Wiley&Sons Inc.,2000:288-308.
[129]王济川,郭志刚.Logistic回归模型—方法与应用.北京:高等教育出版社,2001:14-17.
[130]方积乾,陈和年.医学研究中logistic回归模型的正确应用（一）.中国卫生统计,1993,10(4):54-57.
[131] Kleinbaum DG, Klein M. Logistic regression: a self-learning text.2ndedition.New York:Springer-Verlag Inc.,2002:227-265.
[132]方积乾,陈和年.医学研究中logistic回归模型的正确应用（二）.中国卫生统计,1993,10(5):61-63.
[133] Cheng PE, Liou M, Aston JAD. Likelihood ratio tests with three-way tables.Journal of the American Statistical Association,2010,105:490,740-749.
[134]钟礼杰,高玉堂,金丕焕.Logistic回归模型的拟合优度检验.中国卫生统计,1993,10(3):55-59.
[135] Agresti A著.张淑梅,王瑞,曾莉,译.属性数据分析引论.第2版.北京:高等教育出版社,2007:187-189.
[136]张岩波.对数线性模型的最优模型筛选策略.中国卫生统计,1996,13(6):4-7.
[137]沈其君.SAS统计分析.北京:高等教育出版社,2008:145-164.
[138]余松林,向惠云.重复测量资料分析方法与SAS程序.北京:科学出版社,2004:183-192.
[139] Hardin JW, Hilbe JM. Generalized estimating equations. New York:Chapman&Hall/CRC Press,2003:55-180.
[1] Fitzmaurice G, Davidian M, Verbeke G, et al. Longitudinal data analysis. BocaRaton,FL:Chapman&Hall/CRC Press,2009:1-78.
[2] Zeger ST, Liang KY. Longitudinal data analysis for discrete and continuous out-comes. Biomeics,1986,42:121-130.
[3]刘新胜,崔书克.缺血性中风的中医药研究进展.河南中医,2012,32(1):129-131.
[4]钟海珍,马斌,高颖.缺血性中风病患者发病3个月致残因素初探.中国中医基础医学杂志,2010,16(1):47-50.
[5]马斌,高颖.中风病发病第7天和第14天证候要素演变规律初步研究.辽宁中医杂志,2006,33(12):1561-1563.
[6]王建华.出血性中风、缺血性中风急性期证候演变规律的研究.中国中医急症,2001,10(4):215-217.
[7]张志斌,王永炎,吕爱平,等.论证候要素与证候靶点应证组合辨证.中医杂志,2006,47(7):483-485.
[8]辛喜燕,张华,高颖.缺血性中风急性期痰热证的证候演变特点及其与神经功能缺损程度的关系.辽宁中医杂志,2010,37(9):1644-1646.
[9]鲍远程,杨文明,胡翠平,等.缺血性中风的中医证候特点研究.中医药临床杂志,2012,24(3):189-192.
[10]Nelder JA, Wedderburn RWM. Generalized linear models. Journal of the RoyalStatistical Society,SeriesA,1972,135:370–384.
[11]夏彦,潘晓平,刘元元,等.广义估计方程在临床试验重复测量资料中的应用.现代预防医学,2005,32(5):444-445.
[12]Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalizedestimating equation approach. Biometrics,1988,44(4):1049-1060.
[13]Pan W. Goodness-of-fit tests for GEE with correlated binary data. ScandinavianJournal of Statistics,2002,29:101-110.
[14]Lin DY, Wei LJ, Ying Z. Checking the cox model with cumulative sums ofmartingale-based residuals. Biometrika,1993,80:557-572.
[15]Formann AK, Kohlmann T. Latent class analysis in medical research. StatMethods Med Res,1996,5:179–211.
[16]Magidson J, Vermunt JK. Latent class factor and cluster models, bi-plots and tri-plots and related graphical displays. Sociological Mehtodology,2001,31:223-264.
[17]Collins LM, Lanza ST. Latent class and latent transition analysis: withapplications in the social, behavioral and health sciences. Hoboken,NJ:John Wiley&Sons,2010:3-224.
[18]Heitzler C, Lytle L, Erickson D, et al. Physical activity and sedentary activitypatterns among children and adolescents: a latent class analysis approach. J PhysAct Health,2011,8(4):457-467.
[19]Hagenaars JA, McCutcheon AL. Applied latent class analysis. Cambridge:Cambridge University Press,2002:89-106.
[20]Lin TH, Dayton CM. Model selection information criteria for non-nested latentclass models. Journal of Education and Behavioral Statistics,1997,22(3):249-264.
[21]Goodman LA. On the assignment of individuals to latent classes. SociologicalMethodology,37(1):1-22.
[22]Silverwood RJ, Nitsch D, Pierce M, et al. Characterizing longitudinal patterns ofphysical activity in mid-adulthood using latent class analysis: results from aprospective cohort study. American Journal of Epidemiology,2011,174(12):1406-1415.
[23]Chung H, Lanza ST, Loken E. Latent transition analysis: inference and estimation.Statistics in Medicine,2008,27(11):1834–1854.
[24]Lanza ST, Dziak JJ, Huang L, et al(2011). PROC LCA&PROC LTA User'sGuide(Version1.2.7). University Park: The Methodology Center, Penn State.Retrieved from http://methodology.psu.edu.
[25]Lanza ST, Patrick ME, Maggs JL. Latent transition analysis: benefits of a latentvariable approach to modeling transitions in substance use. J Drug Issues,2010,40(1):93-120.
[26]Lanza ST, Collins LM. A new SAS procedure for latent transition analysis:transitions in dating and sexual risk behavior. Developmental Psychology,2008,44(2):446-456.
[27]Dempster AP, Laird NM, Rubin DB. Maximum likelihood from incomplete datavia EM algorithm(with discussion). Journal of the Royal Statistical Society,SeriesB,1977,39:1–38.
[28]Connell A, Bullock BM, Dishion TJ, et al. Family intervention effects on co-occurring early childhood behavioral and emotional problems: a latent transitionanalysis approach. J Abnorm Child Psychol,2008,36(8):1211-1225.
[29]余民宁.项目反应理论IRT及其应用.台北:心理出版社,2009:1-50.
[30]杜文久.高等项目反应理论.重庆:西南师范大学出版社,2007:52-277.
[31]Embretson SE, Reise SP. Item response theory for psychologists. Mahwah,NewJersey:Lawrence Erlbaum Associates,2000:3-246.
[32]Baker FB. The basics of item response theory.2ndedition.College Park,MD:ERICClearinghouse on Assessment and Evaluation,2001:5-128.
[33]许祖慰.项目反应理论及其在测验中的应用.上海:华东师范大学出版社,1992:18-129.
[34]Baker FB. Methodology reviews: item parameter estimation under the one-, two-,and three-parameter logistic models. Applied Psychological Measurement,1987,11:111-142.
[35]Bock RD, Aitkin M. Marginal maximum likelihood estimation of item parameters:an application of the EM algorithm. Psychometrika,1981,46:443–459.
[36]Kingston NM, Dorans NJ. The analysis of item-ability regression: an exploratoryIRT model fit tool. Applied Psychological Measurement,1985,9:281-288.
[37]Ludlow LH. A strategy for the graphical representation of Rasch model residuals.Educational and Psychological Measurement,1985,45:851-859.
[38]Yen WM. Using simulation results to choose a latent trait model. AppliedPsychological Measurement,1981,5:245-262.
[39]耿晓娟,张军平,高颖,等.缺血性中风病急性期证候变化规律研究.中华中医药杂志,2010,25(9):1485-1487.
[40]杨利,黄燕,蔡业峰,等.1418例中风患者痰瘀证候分布和演变规律探析.辽宁中医杂志,2004,31(6):459-460.
[41]王顺道,杜梦华,解庆凡,等.中风病急性期证候演变规律的研究.中国中医急症,1995,5(3):121-124.
[42]徐俊峰,李保洋,高颖,等.缺血性中风病证候演变模式的研究.天津中医药,2009,26(6):511-514.
[43]谢颖桢,邹忆怀,马洪明,等.中风病急性期中经络和中脏腑证候要素的动态比较.中国中医基础医学杂志,2006,12(10):746-748.
[44]张华.缺血性中风证候演变与神经功能缺损的相关性研究.北京中医药大学学报,2007,30(4):274-278.
[45]刘强,高颖.中风病不同阶段四诊信息的指标聚类分析.中华中医药杂志,2007,22(3):162-164.
[2]刘新胜,崔书克.缺血性中风的中医药研究进展.河南中医,2012,32(1):129-131.
[3]韦桂梅.缺血性中风中医辨证研究近况.云南中医中药杂志,2006,27(3):63-64.
[4]赵国强.缺血性中风药物治疗现状.医学综述,1999,5(4):153-156.
[5]丁自娟.《内经》论治中风.中国中医急症,2010,19(1):111-112.
[6]张填,马涤辉,孙龙.脑卒中患者预后影响因素的分析.医学综述,2007,13(5):385-387.
[7]耿晓娟,张军平,高颖,等.缺血性中风病急性期证候变化规律研究.中华中医药杂志,2010,25(9):1485-1487.
[8]钟海珍,马斌,高颖.缺血性中风病患者发病3个月致残因素初探.中国中医基础医学杂志,2010,16(1):47-50.
[9]马斌,高颖.中风病发病第7天和第14天证候要素演变规律初步研究.辽宁中医杂志,2006,33(12):1561-1563.
[10]王建华.出血性中风、缺血性中风急性期证候演变规律的研究.中国中医急症,2001,10(4):215-217.
[11]郭蕾,王永炎,张志斌.证候概念的诠释.北京中医药大学学报,2003,26(2):5-8.
[12]王永炎,张志斌.再议完善辨证方法体系的几个问题.天津中医药,2007,24(1):1-4.
[13]张志斌,王永炎,吕爱平,等.论证候要素与证候靶点应证组合辨证.中医杂志,2006,47(7):483-485.
[14]刘伍立,欧阳建军,黄博辉.中医文献对中风病的阐述与述评.针灸临床杂志,2006,22(10):5-8.
[15]辛喜燕,张华,高颖.缺血性中风急性期痰热证的证候演变特点及其与神经功能缺损程度的关系.辽宁中医杂志,2010,37(9):1644-1646.
[16]鲍远程,杨文明,胡翠平,等.缺血性中风的中医证候特点研究.中医药临床杂志,2012,24(3):189-192.
[17]杨利,黄燕,蔡业峰,等.1418例中风患者痰瘀证候分布和演变规律探析.辽宁中医杂志,2004,31(6):459-460.
[18]王顺道,杜梦华,解庆凡,等.中风病急性期证候演变规律的研究.中国中医急症,1995,5(3):121-124.
[19]徐俊峰,李保洋,高颖,等.缺血性中风病证候演变模式的研究.天津中医药,2009,26(6):511-514.
[20]谢颖桢,邹忆怀,马洪明,等.中风病急性期中经络和中脏腑证候要素的动态比较.中国中医基础医学杂志,2006,12(10):746-748.
[21]张华.缺血性中风证候演变与神经功能缺损的相关性研究.北京中医药大学学报,2007,30(4):274-278.
[22]刘强,高颖.中风病不同阶段四诊信息的指标聚类分析.中华中医药杂志,2007,22(3):162-164.
[23] Nelder JA, Wedderburn RWM. Generalized linear models. Journal of theRoyal Statistical Society,SeriesA,1972,135:370–384.
[24] Sutradhar BC. An overview on regression models for discrete longitudinalresponses. Statistical Science,2003,18(3):377-393.
[25] Fitzmaurice G, Davidian M, Verbeke G, et al. Longitudinal data analysis.BocaRaton,FL:Chapman&Hall/CRC Press,2009:1-78.
[26] Liang KY, Zeger ST. Longitudinal data analysis using generalized linearmodels. Biometrika,1986,73(1):13-22.
[27] Zeger ST, Liang KY. Longitudinal data analysis for discrete and continuousoutcomes. Biomeics,1986,42:121-130.
[28]高燕宁,蔡文玮,周纪芗.广义估计方程GEE1与纵向资料的回归分析.数理医药学杂志,1994,7(2):118-123.
[29] SAS Institute Inc. SAS/STAT9.2user’s guide. Cary,NC:SAS Institute Inc.,2008:1982-2070.
[30]方积乾.现代医学统计学.北京:人民卫生出版社,2002:25-61.
[31] Liang KY, Zeger SL, Quaqish B. Multivariate regression analysis forcategorical data. Journal of the Royal Statistical Society,Series B,1992,54:3-40.
[32]夏彦,潘晓平,刘元元,等.广义估计方程在临床试验重复测量资料中的应用.现代预防医学,2005,32(5):444-445.
[33] Horton NJ, Lipsitz SR. Review of software to fit generalized estimatingequation regression models. The American Statistician,1999,53:160-169.
[34]朱玉,王静,何倩.广义估计方程在SPSS统计软件中的实现.中国卫生统计,2011,28(2):199-201.
[35]陈启光.纵向研究中重复测量资料的广义估计方程分析.中国卫生统计,1995,12(1):22-25.
[36] Carey VC, Zeger SL, Diggle P. Modelling multivariate binary data withalternating logistic regressions. Biomerika,1993,80(3):517-526.
[37] Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalizedestimating equation approach. Biometrics,1988,44(4):1049-1060.
[38]李运明,封宗超,夏勋,等.广义估计方程在CT显示方法研究数据分析中的应用.现代生物医学进展,2010,10(21):4173-4176.
[39] Akaike H. Factor analysis and the AIC. Psychometrika,1987,52:317–332.
[40] Raftery AE. Bayesian model selection in social research. Sociological Metho-dology,1995,25:111–163.
[41] Park T. A comparison of the generalized estimating equation approach withthe maximum likelihood approach for repeated measurements. Stat Med,1993,12(18):1723-1732.
[42] Pan W. Goodness-of-fit tests for GEE with correlated binary data.Scandinavian Journal of Statistics,2002,29:101-110.
[43] Pan W. Akaike's information criterion in generalized estimating equations.Biometrics,2001,57(1):120-125.
[44] Williamson JM, Lin HM, Barnhart HX. A classification statistic for GEEcategorical response models. Journal of Data Science,2003,1:149-165.
[45] Lin DY, Wei LJ, Ying Z. Checking the cox model with cumulative sums ofmartingale-based residuals. Biometrika,1993,80:557-572.
[46] Lin DY, Wei LJ, Ying Z. Model-checking techniques based on cumulativeresiduals. Biometrics,2002,58:1-12.
[47]罗天娥,赵晋芳,刘桂芬.GENMOD过程和GLIMMIX过程的比较.中国卫生统计,2010,27(2):196-200.
[48]罗天娥,赵晋芳,刘桂芬.累积残差在广义估计方程模型诊断中的应用.中国卫生统计,2009,26(4):387-390.
[49] Lazarsfeld PF. The logical and mathematical foundations of latent structureanalysis. NJ:Princeton University Press,1950:362-412.
[50] Formann AK, Kohlmann T. Latent class analysis in medical research. StatMethods Med Res,1996,5:179–211.
[51]张洁婷,焦璨,张敏强,等.潜在类别分析技术在心理学研究中的应用.心理科学进展,2010,18(12):1991-1998.
[52]申春悌,陈启光,陆岩,等.中医证候要素研究中潜在变量模型的应用.北京中医药大学学报,2010,33(11):725-731.
[53]陈炳为,陈启光,许碧云.潜在变量模型及其在中医证候中的应用概述.中国卫生统计,2009,26(5):535-538.
[54] Collins LM, Lanza ST. Latent class and latent transition analysis: withapplications in the social, behavioral and health sciences. Hoboken,NJ:JohnWiley&Sons,2010:3-224.
[55] Lanza ST, Bary BC. Transitions in drug use among high-risk women: anapplication of latent class and latent transition analysis. Adv Appl Stat Sci,2010,3(2):203-235.
[56] Smith RA, Barclay VC, Smith JLF. Investigating preferences for mosquito-control technologies in mozambique with latent class analysis. Malaria Journal,2011,10:200.
[57] Nylund KL, Asparouhov T, Muthén Bo. Deciding on the number of classes inlatent class analysis and growth mixture modeling: a Monte Carlo simulationstudy. Struct Equ Modeling,2007,14(4):535-569.
[58] Sotres-Alvarez D, Herring AH, Siega-Riz AM. Latent class analysis is usefulto classify pregnant women into dietary patterns. The Journal of Nutrition,2010,140:2253-2259.
[59] Trivedi RB, Ayotte BJ, Thorpe CT, et al. Is there a nonadherent subtype ofhypertensive patient? A latent class analysis approach. Patient Preference andAdherence,2010,4:255-262.
[60] Magidson J, Vermunt JK. Latent class models for clustering: a comparisonwith k-means. CJMR,2002,20:37–44.
[61] Beeber AS, Thorpe JM, Clipp EC. Community-based service use by elderswith dementia and their caregivers: a latent class analysis. Nurs Res,2008,57(5):312–321.
[62] Thorpe JM, Thorpe CT, Kennelty KA, et al. Patterns of perceived barriers tomedical care in older adults: a latent class analysis. BMC Health ServicesResearch,2011,11:181.
[63] Lanza ST, Collins LM, Lemmon DR, et al. PROC LCA: a SAS procedure forlatent class analysis. Structural Equation Modeling,2007,14(4):671-694.
[64] Magidson J, Vermunt JK. Latent class factor and cluster models, bi-plots andtri-plots and related graphical displays. Sociological Mehtodology,2001,31:223-264.
[65]陈炳为,郑水龙,刘沛,等.潜在分类模型在无金标准诊断试验中的评价.中国卫生统计,2010,27(1):11-16.
[66]张岩波.潜变量分析.北京:高等教育出版社,2009:220-246.
[67] Heitzler C, Lytle L, Erickson D, et al. Physical activity and sedentary activitypatterns among children and adolescents: a latent class analysis approach. JPhys Act Health,2011,8(4):457-467.
[68] Boyko EJ, Doheny RA, McNeely MJ, et al. Latent class analysis of themetabolic syndrome. Diabetes Res Clin Pract,2010,89(1):88-93.
[69]郭小玲,裴磊磊,张岩波.潜在类别模型及数据模拟分析.数理医药学杂志,2009,22(6):631-635.
[70] Lanza ST, Dziak JJ, Huang L, et al(2011). PROC LCA&PROC LTA User'sGuide(Version1.2.7). University Park: The Methodology Center, Penn State.Retrieved from http://methodology.psu.edu.
[71]邱皓政.潜在类别模型的原理与技术.北京:教育科学出版社,2008:1-47.
[72] Hagenaars JA, McCutcheon AL. Applied latent class analysis. Cambridge:Cambridge University Press,2002:89-106.
[73] Dean N, Raftery AE. Latent class analysis variable selection. Ann Inst StatMath,2010,62(1):11-35.
[74] Goodman LA. Exploratory latent structure analysis using both identifiable andunidentifiable models. Biometrika,1974,61(2):215-231.
[75] Formann AK. Linear logistic latent class analysis for polytomous data. Journalof the American Statistical Association,1992,87:476-486.
[76] Lin TH, Dayton CM. Model selection information criteria for non-nestedlatent class models. Journal of Education and Behavioral Statistics,1997,22(3):249-264.
[77] Silverwood RJ, Nitsch D, Pierce M, et al. Characterizing longitudinal patternsof physical activity in mid-adulthood using latent class analysis: results from aprospective cohort study. American Journal of Epidemiology,2011,174(12):1406-1415.
[78] Goodman LA. On the assignment of individuals to latent classes. SociologicalMethodology,2007,37(1):1-22.
[79]孟灿,武俊青,李玉艳,等.潜类别分析原理及其在聚类分析中的应用.中国卫生统计,2010,27(3):237-239.
[80] Chung H, Lanza ST, Loken E. Latent transition analysis: inference andestimation. Statistics in Medicine,2008,27(11):1834–1854.
[81] Lanza ST, Patrick ME, Maggs JL. Latent transition analysis: benefits of alatent variable approach to modeling transitions in substance use. J DrugIssues,2010,40(1):93-120.
[82] Lanza ST, Collins LM. A new SAS procedure for latent transition analysis:transitions in dating and sexual risk behavior. Developmental Psychology,2008,44(2):446-456.
[83] Roberts TJ, Ward SE. Using latent transition analysis in nursing research toexplore change over time. Nurs Res,2011,60(1):73-79.
[84] Dempster AP, Laird NM, Rubin DB. Maximum likelihood from incompletedata via EM algorithm(with discussion). Journal of the Royal StatisticalSociety,Series B,1977,39:1–38.
[85] Connell A, Bullock BM, Dishion TJ, et al. Family intervention effects on co-occurring early childhood behavioral and emotional problems: a latenttransition analysis approach. J Abnorm Child Psychol,2008,36(8):1211-1225.
[86] Alagumalai S, Curtis DD, Hungi N. Applied rasch measurement: a book ofexemplars. Netherlands:Springer,2005:1-14.
[87] Novick MR. The axioms and principal results of classical test theory. Journalof Mathematical Psychology,1966,3:1–18.
[88]余民宁.项目反应理论IRT及其应用.台北:心理出版社,2009:1-50.
[89]漆书青,戴海崎.项目反应理论及其应用研究.南昌:江西高校出版社,1992:129-215.
[90] Embretson SE, Reise SP. Item response theory for psychologists. Mahwah,New Jersey:Lawrence Erlbaum Associates,2000:3-246.
[91] Bock RD. Estimating item parameters and latent ability when responses arescored in two or more nominal categories. Psychometrika,1972,37:29–51.
[92] Mellenbergh GJ. Conceptual notes on models for discrete polytomous itemresponses. Applied Psychological Measurement,1995,19:91–100.
[93]高怀勇,金桂林.项目反应理论及其在计算机自适应测试中的应用.西华师范大学学报（自然科学版）,2008,29(1):84-88.
[94] Kosinski M, Bayliss MS, Bjorner JB, et al. A six-item short-form survey formeasuring headache impact: the HIT-6. Quality of Life Research,2003,12(8):963-974.
[95] Green DR, Yen WM, Burket GR. Experience in the application of itemresponse theory in test construction. Applied Measurement in Education,1989,2(4):297-312.
[96] Weiss DJ. Adaptive testing by computer. Journal of Consulting and ClinicalPsychology,1985,53:774-789.
[97] Ware JE, Kosinski M, Bjorner JB, et al. Applications of computerizedadaptive testing (CAT) to the assessment of headache impact. Quality of LifeResearch,2003,12(9):935-952.
[98]查尔斯L赫林.项目反应理论—在心理测量中的应用.武汉:湖北教育出版社,1990:14-82.
[99] Van der Linden WJ, Hambleton RK. Handbook of modern item responsetheory. New York:Springer-Verlag,1997:1-28.
[100]陈炳为,许碧云,陈启光,等.两分类项目反应理论在中医证候中的应用.中国卫生统计,2011,28(1):16-18,21.
[101] Lai JS, Cella D, Chang CH, et al. Item banking to improve, shorten andcomputerize self-reported fatigue: an illustration of steps to create a core itembank from the FACIT-Fatigue Scale. Quality of Life Research,2003,12(5):485-501.
[102] Robitail S, Erhart M, Tebe C, et al. Person fit across European countries:results from the kidscreen field survey. Quality of Life Research,2004,13(9):1547.
[103]许祖慰.项目反应理论及其在测验中的应用.上海:华东师范大学出版社,1992:18-129.
[104] McDonald RP. A basis for multidimensional item response theory. AppliedPsychological Measurement,2000,24:99-114.
[105] Reckase MD. The past and future of multidimensional item response theory.Applied Psychological Measurement,1997,21(1):25-36.
[106] Reckase MD, Ackerman TA, Carlson JE. Building a unidimensional test usingmultidimensional items. Journal of Educational Measurement,1988,25:193-203.
[107]林岳卿,方积乾.多维IRT与单维IRT在多维量表中应用的差异.中国卫生统计,2011,28(3):226-228.
[108] Reeves BB, Hays RD, Bjorner JB, et al. Psychometric evaluation andcalibration of health-related quality of life item banks: plans for the patient-reported outcomes measurement information systems(PROMIS). Med Care,2007,45:22-31.
[109] Davier MV, Carstensen CH. Multivariate and mixture distribution raschmodels: extensions and applications. New York:Springer,2007:1-55.
[110]余嘉元.项目反应理论及其应用.南京:江苏教育出版社,1992:1-157.
[111]杜文久.高等项目反应理论.重庆:西南师范大学出版社,2007:52-277.
[112] Muthén LK, Muthén BO. Mplus user’s guide.6thedition. Los Angeles,CA:Muthén&Muthén,2010:60-61.
[113] Baker FB. The basics of item response theory.2ndedition. College Park,MD:ERIC Clearinghouse on Assessment and Evaluation,2001:5-128.
[114] Baker FB. Methodology reviews: item parameter estimation under the one-,two-, and three-parameter logistic models. Applied Psychological Measure-ment,1987,11:111-142.
[115] Bock RD, Aitkin M. Marginal maximum likelihood estimation of itemparameters: an application of the EM algorithm. Psychometrika,1981,46:443–459.
[116] Kingston NM, Dorans NJ. The analysis of item-ability regression: anexploratory IRT model fit tool. Applied Psychological Measurement,1985,9:281-288.
[117] Ludlow LH. A strategy for the graphical representation of Rasch modelresiduals. Educational and Psychological Measurement,1985,45:851-859.
[118] Yen WM. Using simulation results to choose a latent trait model. AppliedPsychological Measurement,1981,5:245-262.
[119]胡良平.医学统计学—应用三型理论分析定量资料与定性资料.北京:人民军医出版社,2009:325-393.
[120] Cox DR. The analysis of binary data. London:Methuen,1970:14-29.
[121]金丕焕.医用统计方法.第2版.上海:复旦大学出版社,2003:332-343,355-365.
[122]胡良平.医学统计学—运用三型理论进行现代回归分析.北京:人民军医出版社,2010:158-174,203-220.
[123]胡良平.SAS统计分析教程.北京:电子工业出版社,2010:183-191,332-343.
[124]张家放.医用多元统计方法.武汉:华中科技大学出版社,2002:127-157.
[125]方积乾.卫生统计学.第5版.北京:人民卫生出版社,2007:356-367.
[126]陈峰.医用多元统计分析方法.第2版.北京:中国统计出版社,2007:82-113.
[127]柳青.多元统计分册—中国医学统计百科全书.北京:人民卫生出版社,2004:218-221.
[128] Hosmer DW, Lemeshow S. Applied logistic regression.2ndedition.John Wiley&Sons Inc.,2000:288-308.
[129]王济川,郭志刚.Logistic回归模型—方法与应用.北京:高等教育出版社,2001:14-17.
[130]方积乾,陈和年.医学研究中logistic回归模型的正确应用（一）.中国卫生统计,1993,10(4):54-57.
[131] Kleinbaum DG, Klein M. Logistic regression: a self-learning text.2ndedition.New York:Springer-Verlag Inc.,2002:227-265.
[132]方积乾,陈和年.医学研究中logistic回归模型的正确应用（二）.中国卫生统计,1993,10(5):61-63.
[133] Cheng PE, Liou M, Aston JAD. Likelihood ratio tests with three-way tables.Journal of the American Statistical Association,2010,105:490,740-749.
[134]钟礼杰,高玉堂,金丕焕.Logistic回归模型的拟合优度检验.中国卫生统计,1993,10(3):55-59.
[135] Agresti A著.张淑梅,王瑞,曾莉,译.属性数据分析引论.第2版.北京:高等教育出版社,2007:187-189.
[136]张岩波.对数线性模型的最优模型筛选策略.中国卫生统计,1996,13(6):4-7.
[137]沈其君.SAS统计分析.北京:高等教育出版社,2008:145-164.
[138]余松林,向惠云.重复测量资料分析方法与SAS程序.北京:科学出版社,2004:183-192.
[139] Hardin JW, Hilbe JM. Generalized estimating equations. New York:Chapman&Hall/CRC Press,2003:55-180.
[1] Fitzmaurice G, Davidian M, Verbeke G, et al. Longitudinal data analysis. BocaRaton,FL:Chapman&Hall/CRC Press,2009:1-78.
[2] Zeger ST, Liang KY. Longitudinal data analysis for discrete and continuous out-comes. Biomeics,1986,42:121-130.
[3]刘新胜,崔书克.缺血性中风的中医药研究进展.河南中医,2012,32(1):129-131.
[4]钟海珍,马斌,高颖.缺血性中风病患者发病3个月致残因素初探.中国中医基础医学杂志,2010,16(1):47-50.
[5]马斌,高颖.中风病发病第7天和第14天证候要素演变规律初步研究.辽宁中医杂志,2006,33(12):1561-1563.
[6]王建华.出血性中风、缺血性中风急性期证候演变规律的研究.中国中医急症,2001,10(4):215-217.
[7]张志斌,王永炎,吕爱平,等.论证候要素与证候靶点应证组合辨证.中医杂志,2006,47(7):483-485.
[8]辛喜燕,张华,高颖.缺血性中风急性期痰热证的证候演变特点及其与神经功能缺损程度的关系.辽宁中医杂志,2010,37(9):1644-1646.
[9]鲍远程,杨文明,胡翠平,等.缺血性中风的中医证候特点研究.中医药临床杂志,2012,24(3):189-192.
[10]Nelder JA, Wedderburn RWM. Generalized linear models. Journal of the RoyalStatistical Society,SeriesA,1972,135:370–384.
[11]夏彦,潘晓平,刘元元,等.广义估计方程在临床试验重复测量资料中的应用.现代预防医学,2005,32(5):444-445.
[12]Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalizedestimating equation approach. Biometrics,1988,44(4):1049-1060.
[13]Pan W. Goodness-of-fit tests for GEE with correlated binary data. ScandinavianJournal of Statistics,2002,29:101-110.
[14]Lin DY, Wei LJ, Ying Z. Checking the cox model with cumulative sums ofmartingale-based residuals. Biometrika,1993,80:557-572.
[15]Formann AK, Kohlmann T. Latent class analysis in medical research. StatMethods Med Res,1996,5:179–211.
[16]Magidson J, Vermunt JK. Latent class factor and cluster models, bi-plots and tri-plots and related graphical displays. Sociological Mehtodology,2001,31:223-264.
[17]Collins LM, Lanza ST. Latent class and latent transition analysis: withapplications in the social, behavioral and health sciences. Hoboken,NJ:John Wiley&Sons,2010:3-224.
[18]Heitzler C, Lytle L, Erickson D, et al. Physical activity and sedentary activitypatterns among children and adolescents: a latent class analysis approach. J PhysAct Health,2011,8(4):457-467.
[19]Hagenaars JA, McCutcheon AL. Applied latent class analysis. Cambridge:Cambridge University Press,2002:89-106.
[20]Lin TH, Dayton CM. Model selection information criteria for non-nested latentclass models. Journal of Education and Behavioral Statistics,1997,22(3):249-264.
[21]Goodman LA. On the assignment of individuals to latent classes. SociologicalMethodology,37(1):1-22.
[22]Silverwood RJ, Nitsch D, Pierce M, et al. Characterizing longitudinal patterns ofphysical activity in mid-adulthood using latent class analysis: results from aprospective cohort study. American Journal of Epidemiology,2011,174(12):1406-1415.
[23]Chung H, Lanza ST, Loken E. Latent transition analysis: inference and estimation.Statistics in Medicine,2008,27(11):1834–1854.
[24]Lanza ST, Dziak JJ, Huang L, et al(2011). PROC LCA&PROC LTA User'sGuide(Version1.2.7). University Park: The Methodology Center, Penn State.Retrieved from http://methodology.psu.edu.
[25]Lanza ST, Patrick ME, Maggs JL. Latent transition analysis: benefits of a latentvariable approach to modeling transitions in substance use. J Drug Issues,2010,40(1):93-120.
[26]Lanza ST, Collins LM. A new SAS procedure for latent transition analysis:transitions in dating and sexual risk behavior. Developmental Psychology,2008,44(2):446-456.
[27]Dempster AP, Laird NM, Rubin DB. Maximum likelihood from incomplete datavia EM algorithm(with discussion). Journal of the Royal Statistical Society,SeriesB,1977,39:1–38.
[28]Connell A, Bullock BM, Dishion TJ, et al. Family intervention effects on co-occurring early childhood behavioral and emotional problems: a latent transitionanalysis approach. J Abnorm Child Psychol,2008,36(8):1211-1225.
[29]余民宁.项目反应理论IRT及其应用.台北:心理出版社,2009:1-50.
[30]杜文久.高等项目反应理论.重庆:西南师范大学出版社,2007:52-277.
[31]Embretson SE, Reise SP. Item response theory for psychologists. Mahwah,NewJersey:Lawrence Erlbaum Associates,2000:3-246.
[32]Baker FB. The basics of item response theory.2ndedition.College Park,MD:ERICClearinghouse on Assessment and Evaluation,2001:5-128.
[33]许祖慰.项目反应理论及其在测验中的应用.上海:华东师范大学出版社,1992:18-129.
[34]Baker FB. Methodology reviews: item parameter estimation under the one-, two-,and three-parameter logistic models. Applied Psychological Measurement,1987,11:111-142.
[35]Bock RD, Aitkin M. Marginal maximum likelihood estimation of item parameters:an application of the EM algorithm. Psychometrika,1981,46:443–459.
[36]Kingston NM, Dorans NJ. The analysis of item-ability regression: an exploratoryIRT model fit tool. Applied Psychological Measurement,1985,9:281-288.
[37]Ludlow LH. A strategy for the graphical representation of Rasch model residuals.Educational and Psychological Measurement,1985,45:851-859.
[38]Yen WM. Using simulation results to choose a latent trait model. AppliedPsychological Measurement,1981,5:245-262.
[39]耿晓娟,张军平,高颖,等.缺血性中风病急性期证候变化规律研究.中华中医药杂志,2010,25(9):1485-1487.
[40]杨利,黄燕,蔡业峰,等.1418例中风患者痰瘀证候分布和演变规律探析.辽宁中医杂志,2004,31(6):459-460.
[41]王顺道,杜梦华,解庆凡,等.中风病急性期证候演变规律的研究.中国中医急症,1995,5(3):121-124.
[42]徐俊峰,李保洋,高颖,等.缺血性中风病证候演变模式的研究.天津中医药,2009,26(6):511-514.
[43]谢颖桢,邹忆怀,马洪明,等.中风病急性期中经络和中脏腑证候要素的动态比较.中国中医基础医学杂志,2006,12(10):746-748.
[44]张华.缺血性中风证候演变与神经功能缺损的相关性研究.北京中医药大学学报,2007,30(4):274-278.
[45]刘强,高颖.中风病不同阶段四诊信息的指标聚类分析.中华中医药杂志,2007,22(3):162-164.