【作者】 纪荣芳；

【导师】 牛建昭； 许树强；

【作者基本信息】 北京中医药大学 ， 中西医结合基础， 2007， 博士

【摘要】 目的主要有以下三点：(1)探讨国内老年痴呆临床科研近况，为未来的研究决策提供参考依据。(2)对中西医治疗老年痴呆近况进行全面、系统的定性评估，同时对符合条件的研究进行定量的比较分析，为临床决策提供依据。(3)对治疗健忘或痴呆的中药方剂进行用药规律分析，为临床用药提供参考。方法循证医学meta分析实验部份研究方法可分为文献收集及筛选和文献定性及定量分析两个步骤。文献收集及筛选乃是从中国知网(http://www.cnki.net)及万方数据库(http://www.wanfangdata.com.cn)中搜寻，键入关键词“老年性痴呆”或”阿尔茨海默症”或“阿尔茨海默病”或“老年呆病”。入选标准为：(1)中文文献(2)发表于1994-2004年间(3)实验设计为以中药治疗，西药(包括安慰剂、维生素)为对照。包括试验性研究及非随机和观察性的比较性研究。(4)各文献研究的问题和方法相似，都为关于中西药对老人痴呆症疗效研究。(5)研究对象被清楚的确诊为痴呆症。总共收集2403篇文章，从中筛选出40个入选研究。文献之定性分析部份乃将入选文献按研究对象、总数、人口学特征、使用方剂、对照药物及测量结果等进行归纳总结，并综述、探讨、比较中西医对阿尔茨海默病的治法及用药：定量分析部份则是将入选文献之研究先用Excel建立数据库，再将入选文献实施同质型检验(homogeneity test)、meta分析(meta analysis)及敏感性分析(sensitivity analysis)。接着并评估发表性偏倚(publication bias)。中医药数据挖掘实验研究方法分为数据收集及数据挖掘两个步骤，数据收集部份乃从中国方剂数据库中，在主治字段中，以痴呆或健忘为关健词搜索，将所得的方剂、药物组成先作数据预处理之后，再用Excel建立数据库。结果共选出方剂197首。数据挖掘部份除描述性分析之外，本研究更采用购物蓝分析(market basket analysis)、因子分析(factor analysis)、聚类分析(cluster analysis)、多元尺度法(multidimentional scaling)等数据挖掘技术来探讨方剂配伍。购物蓝分析部分乃采用SAS 9.1 for Windows中的enterprise minet模块为挖掘工具。其它部分的研究则使用利用美国犹他州(Utah)，Kaysville的NCSS公司生产的NCSS(number cruncher statistical systems) software来挖掘。结果循证医学meta分析实验部份发现：从1997年到2004年止入选的这40个研究中，整体看起来，质量不高。其中有3个研究之病例年龄未明确表示，有5个研究未标明治疗天数，其它有标明者，治疗天数从15天到240天不等。实验设计方面，共有32个属于试验性研究，8个属于及非随机和观察性的比较性研究。疗效指标方面，有一个研究未标明。对于老年人的慢性病治疗来说，治疗天数太短便难以看出疗效。疗效指标未标明，便难有说服性。依Jadad量表来评随机对照试验发现，最佳总分为2分，共有7个(7／32=22％)，总分1分的有25个(25／32=78％)。尤其是双盲法的使用及退出与失访部分：仅有4个提到双盲法，另有3篇提到用单盲法，全部皆无提到退出与失访部分。西医对照药物来看，主要分为乙酰胆碱酯酶(AchE)抑制剂(以使用多奈哌齐4组最多)、脑循环改善和糖代谢促进剂(以脑复康17组最多)和其它三类。异质性检验显示：无论选择比值比或危险差当效应量指标，P值皆＜0.0001，提示组间异质性是具有统计学上意义的。Radial图显示：4，14，16，19，26，与35组为孤立点(outlier)，可能就是本研究组间异质性(heterogeneity)的最大来源。随后用敏感性分析显示第14项研究对meta分析的结果影响最大，应是异质性的最大来源。最后，采用随机效应模型来计算，结果合并比数比为3.24，95％的可信区间为2.3-4.56。接着分别依出版年份、控制组种类及不同实验设计做亚组分析。结果显示出版于2002年以前，OR=4.76(2.77～8.18)，出版于2002年及其以后，OR=2.11(1.59～2.78)，提示2002年以后中药对西药的疗效比值下降。依控制组的4个亚组做分析时，有3个亚组都是中药疗效优于西药，只有和神经传导介质类药物相比时，比数比为1.20，95％可信区间为0.73到1.97之间，不具统计学上之意义。可能的解释是目前西医对老年痴呆症是以神经传导介质类药物为主流，其疗效比非神经传导介质类药物佳。中药治疗与其相较，优势相对减弱。不同实验设计做亚组分析时发现：试验性研究，中药疗效比西药疗效合并比数比为2.92，95％可信区间为2.10到4.07之间：比较性研究，中药疗效比西药疗效比数比为4.38，95％可信区间为1.62到11.85之间，皆具统计学上之意义。中医药数据挖掘实验部份发现：中医治疗痴呆的药物种类分布广泛。从用药频次看来，前三多者为补虚药，占所有使用频次的29％，安神药次之，占所有使用频次的14.8％，清热药再次之，占所有使用频次的10.3％。而补虚药之核心药物为人参、当归及熟地；安神药之核心药物为远志、酸枣仁及朱砂；清热药之核心药物为甘草、生地及黄连。整体来看，出现频次大于等于25次(=12.7％)之中药，按出现频次由多自少依次为：人参、远志、茯神、茯苓、当归、甘草、熟地、酸枣仁、麦冬、石菖蒲、朱砂、五味子、黄耆、桂、柏子仁、桑寄生、生地、龙骨、川牛膝、山药、菟丝子、天门冬、防风等共23味。其中除远志、石菖蒲以辛、苦为主味之外，都以甘味为主。从药性上来看：没有大寒的药物，寒性药物有：朱砂、生地、天门冬：微寒药物有：甘草、麦冬：平性药物有：茯神、茯苓、酸枣仁、柏子仁、桑寄生、龙骨、川牛膝、山药：微温的有：人参、远志、熟地、黄耆、菟丝子、防风：温性的有：当归、血味子、石菖蒲：大热的药物只有桂一种。从归经看：心、肝、脾、胃、肺、肾各经都有分布，但以心、肾、脾三经为主。除熟地、黄耆、川牛膝、山药、桑寄生、菟丝子、天门冬、防风外都入心经。入胃经的药物最少只有甘草、麦冬、石菖蒲、生地、天冬等五种。以药物类别来看：此23味中药依然以补虚药9种居先，安神药5种次之。从二味药购物篮分析表看来，交易数最大的前七种关联规则为：茯苓(?)远志，茯神(?)远志，茯苓(?)人参，石菖蒲(?)远志，熟地(?)远志，酸枣仁(?)远志，酸枣仁(?)人参。可见主要药物配伍是以安神类，尤其是以远志、酸枣仁为主。其基本思路是从心不藏神来论治。因子分析发现用四个因子解释较恰当，累积解释能力为68.4％。其中，第一因子含菟丝子、川牛膝、五味子、熟地、柏子仁、山药：第二因子含桂、酸枣仁、生地、防风、石菖蒲；第三因子含黄蓍、桑寄生、人参、甘草；第四因子含天冬、麦冬、远志，茯神。特别注意的是远志与茯神同属于一共同因子，与关联分析产生之规则相应。聚类分析选择将此数据分为5个聚类但由于平均silhouette值仅为0.05，找出规则之说服力较薄弱。多元尺度分析结果显示，多元尺度分析图上群聚现象不强，但用三个维度模型来建模是可接受的。结论循证医学meta分析实验部份：经过文献检索之后发现，符合纳入和排除标准的试验不多，况且入选之随机试验和(或)半随机试验质量不高，比较性研究之证据等级也不高。在此基础上meta分析结果外在效度较低。由于此推论是在质量不高的文献上meta分析的结果，要强化此观点仍需要研究质量良好的随机对照试验之科研成果来支持。解决之道是强化科研设计质量并借由选取同种疾病中具有相同证候的患者一起研究，以增加资料同质性，并借由严格的入选标准方可客观地评价中医药的疗效，用循证医学的方法和原则来研究方可得到国际认同。中医药数据挖掘实验部份：根据用药频次统计及数据挖掘发现中医治疗健忘或痴呆之基本思路是从心不藏神来论治，中医认为此等病症病理上属本虚标实，故尤重补虚药、安神药及清热药之使用。更多还原

【Abstract】 This research selected Alzheimer’ s disease- one of the mostchallenging and important health issues for the elders in the 21stcentury-as the subject to deal with. The rationale for this research ismainly six-fold: (1) Aging society is one of the major public health issue(2) the unmet need in aging society: to improve the quality of life ofthe elders (3) Alzheimer’ s disease is one of the major health threatsto the elders in China (4)Many researches showed that the treatment effectof Traditional Chinese Medicine (TCM) is much better than WesternMedicine (5) If these treatment effects derived from various studies couldbe synthesized, that will be conducive to clinical decision-making. (6)Using two internationally- recognized research tools: meta analysis anddata mining to exp]ore Alzheimer’ s treatment is an innovative way ofresearch on TCM, which will help TCM with internalization.The purpose of this research was mainly three-fold: to get anunderstanding of current researches conducted on Alzheimer’ s disease inChina, to systematically evaluate and compare Alzheimer’ s treatmentdelivered by TCM with that by Western Medicine and also conductquantitative analysis on qualified researches, to discover meaningfulhidden patterns of Chinese drug compatibility used for amnesia and/ordementias.This dissertation contains three parts, part one was literaturereview, part two and part three discussed Alzheimer’ s disease mainly fromthe viewpoint of meta analysis and data mining, respectively. There werefour main topics reviewed in part one, including (1) The current advancesin evidence-based medicine (2) The current advances in data mining (3)Alzheimer’ s disease-East meets West (4) Treatment of Alzheimer’ sdisease based on invigoration if both Spleen and KidneyThe data source in meta-analysis of Alzheimer’ s treatment came fromthe following two data base: cnki (http://www.cnki.net)and wanfangdata(http://www.wanfangdata.com.cn). Inclusion criteria were: (1) Chineseliterature (2) Published between year 1994 and year 2004 (3) Using TCMas treatment and Western Medicine as control (4) Similar research purposeand methodology (5) Subjects were diagnosed as Alzheimer’ s disease (6)Objective outcome measurement (7) Being statistically able to besynthesized by meta-analysis Only 40 out of 2403 studies met the inclusion criteria. The other data source used in data mining experimentcame from Chinese Decoction data base. After using dementia or amnesiaas keyword, 197 decoctions were collected and processed for data mining.Findings in meta-analysis of Alzheimer’ s treatment: From year 1997to year 2004, there were forty qualified studies. All in all, the studyquality deserved improving. Among those studies, there were three studieswithout giving patients’ ages, five without marking days of treatment.For those studies with days of treatment, the treatment duration wasbetween 15 to 240 days. There was also one study without describing outcomemeasurement tool. In evaluating the treatment effects of chronic disease,short days of treatment possibly cannot warrant effective. When assessedby Jadad score, 7 out of 40 studies got a score of 2 points, 25 out of40 studies got 1 point, and the rest 8 studies scored zero. None of thesestudies obtained a Jadad score of more than 2. Especially, in blindingof the patients and the description of losses to follow-up items, only4 studies mentioned double-blinding and three mentioned single-blindingmethod. None described attrition. Regarding control groups, they cangenerally be categorized into three groups: cholinesterase inhibitor,intelligence booster and CNS stimulant and others. In cholinesteraseinhibitor control group, there were 4 studies using Donepezil as control.In intelligence booster and CNS stimulant group, there were 17 studiesusing Piracetam as control. These two drugs were the most frequently usedmedication in their own categories.Heterogeneity test showed statistically significant with P valueless than 0.0001, no matter odds ratio or risk difference was chosen asoutcome measurement. Radial plot revealed the following studies asoutliers: study 4, studyl4, studyl6, studyl9, study 26 and study 35.Sensitivity analysis also showed that study 14 was the largest sourceof heterogeneity.Finally, odds ratio were pooled using random effect model. The resultsshowed odds ratio=3.24, 95% confidence interval was between 2.30 and4.56.In order to explore possible sources of heterogeneity, subgroup analysis was also conducted by different control groups and publicationdate, respectively. The cutoff point for publication date was year 2002(included) and four different control groups, includingneurotransmitters, non-neurotransmitters, placebo and combined therapywere investigated. The results showed OR=4.76 with 95% confidenceinterval between 2.77 and 8.18 and OR=2.11 with 95% confidence interval between 1.59 and 2.78 for earlier publication and later publicationgroups, respectively. This implies that after year 2002, the advantageof Traditional Chinese Medicine is losing. Another subgroup analysis bydifferent control groups showed TCM was no better than neurotransmittersgroup with OR=1.2, 95% confidence interval between 0.73 and 1.97, whichwas statistically insignificant. However, TCM showed more advantageousthan non-neurotransmitters, placebo and combined therapy. Possibleexplanation for this finding is because the therapeutic effect ofCholinesterase inhibitors, the mainstream of current Alzheimer’ stherapy, is generally better than neuro-protective approaches,therefore, when compared with TCM, there was no significant difference.Findings in data mining experiment on Alzheimer’ s treatment:According to theories of TCM, there are even more approaches toAlzheimer’ s treatment. The commonly used drugs covered many differentcategories. In decreasing order of frequency used, they are tonics, whichaccounted for 29% of total frequency, followed by tranquilizers, whichaccounted for 14.8% and antipyretics, which accounted for 10.3%. Thefrequently used drugs in tonics were Renshen, Danggui and Shudihuang. Intranquilizers category, they were Yuanzhi, Suanzaoren and Zhusha. Inantipyretics, they were Gancao, Dihuang and Huanglian. Overall, frequencydistribution for top 23 (=12.7%) prescription drugs, in decreasingorder of frequency, were Renshen, Yuanzhi, Fushen, Fuling, Danggui,Gancao, Shudihuang, Suanzaoren, Maidong, Shichangpu, Zhusha, Wuweizi,Hungqi, Gui, Baiziren, Sangjisheng, Shengdi, Longgu, Chuanniuxi, Shanyao,Tusizi, Tiandong and Fangfeng.They were generally with sweet flavor, except for Yuanzhi andShichangpu which were pungent and bitter in taste. There were no drugsof extremely cold character. The drugs of cold character were Zhusha,Shengdi and Tiandong; the drugs of slightly cold character were Gancaoand Maidong; the drugs of mild nature were Fushen, Fuling, Suanzaoren,Baiziren, Sangjisheng, Longgu, Chuanniuxi and Shanyao; the drugs oflukewarm character were Renshen, Yuanzhi, Shudihuang, Hungqi, Tusizi andFangfeng; the drugs of warm character were Danggui, Wuweizi and Shichangpu;Only Gui belong to drug of extremely warm character. Regarding channeltropism, all drugs’ therapeutic effects can cover heart, liver, spleen,stomach, lung and kidney channels, but mainly on heart, kidney and spleenchannels. All drugs except Shudihuang, nungqi, Chuanniuxi,Shanyao, Sangjisheng, Tusizi, Tiandong, and Fangfeng could havetherapeutic effects on heart channel. Only Gancao, Maidong, Shichangpu,Shengdi and Tiandong could be distributed to stomach channel. Most ofthese frequently used drugs belong to tonics and tranquilizers. Among them, 9 drugs belong to tonics and 5 drugs belong to tranquilizers.Association rule mining section showed the following mostfrequently used associations, in decreasing order of transaction, were:Fuling(?)Yuanzhi, Fushen(?)Yuanzhi, Fuling(?)Renshen, Shichangpu(?)Yuanzhi, Shudihuang(?) Yuanzhi, Suanzaoren(?)Yuanzhi,Suanzaoren(?)Renshen. This implies that the major component of drugcompatibility went to tranquilizers, especially Suanzaoren and Yuanzhi.Inferring from the above association rules, TCM mostly treat Alzheimer’ sdisease as a kind of heart disease. Because based upon one TCM theory:the heart dominates all mental activities.In factor analysis section, Kaiser’ s criteria were applied todrop factors whose eigenvalues less than one and four factors wereretained. The cumulative percentage of the total variation in thevariables accounted for by these four factors amounted to 68.4%. The firstfactor included Tusizi, Chuanniuxi, Wuweizi, Shudihuang, Baiziren andShanyao. The second factor included Gui, Suanzaoren, Shengdi, Fangfengand Shichangpu. The third factor included Hungqi, Sangjisheng, Renshenand Gancao. The fourth factor included Tiandong, Maidong, Yuanzhi andFushen.Five clusters were determined in cluster analysis section. However,the maximum average silhouette equaled to 0.05, indicating no substantialstructure had been found.Finally, in multidimensional scaling section, the results showedno obvious clustering but the goodness-of-fit in 3 dimensional model isacceptable.更多还原