1、系統(tǒng)評價和Meta-分析1We can see further than our forbearers because we stand on the shoulders of Giants.2問題針對一個問題，有多項(xiàng)研究報(bào)道 研究方法相同或不同 結(jié)果一致或不一致如何對這一問題的研究現(xiàn)狀全面了解、客觀分析，指導(dǎo)于臨床實(shí)踐？文獻(xiàn)綜述： 帶有很大程度的主觀因素 文獻(xiàn)質(zhì)量控制不嚴(yán)格 收集的文獻(xiàn)不全面 不能進(jìn)行定量的統(tǒng)計(jì)分析 有沒有更科學(xué)的方法？ 3系統(tǒng)評價 (systematic review，SR)Meta-分析 (meta-analysis)4系統(tǒng)評價Overview, systematic o

2、verview, pooling project.系統(tǒng)評價是一種臨床研究方法，是全面收集相關(guān)的所有臨床研究，并逐個進(jìn)行嚴(yán)格評價和分析，必要時進(jìn)行定量合成的統(tǒng)計(jì)學(xué)處理，得出綜合結(jié)論的過程?！跋到y(tǒng)”和“評價”是Systematic review的重要特點(diǎn)， 與一般的綜述有本質(zhì)不同(在權(quán)衡研究的取舍時往往帶有選擇性和主觀性)。 5概 述60年代開始，在醫(yī)學(xué)文獻(xiàn)中，陸續(xù)出現(xiàn)了對多個獨(dú)立研究的統(tǒng)計(jì)量進(jìn)行合并的報(bào)道。1976年G.V.Glass首先將合并統(tǒng)計(jì)量將文獻(xiàn)進(jìn)行分析的方法稱為“Meta-analysis”。80年代末傳入我國，中文譯名有薈萃分析、二次分析、匯總分析、集成分析等。7Meta-分析 M

3、eta-分析是依靠收集已經(jīng)或未發(fā)表的具有某一可比特性的文獻(xiàn)，應(yīng)用特定的設(shè)計(jì)和統(tǒng)計(jì)學(xué)方法進(jìn)行分析與綜合評價，使有可能對具有不同設(shè)計(jì)方法及不同病例數(shù)的研究結(jié)果進(jìn)行綜合比較。 Meta-分析是系統(tǒng)評價的一種，是一種研究過程。 狹義：Meta-分析只是一種定量合成的統(tǒng)計(jì)處理方法。 8系統(tǒng)評價/Meta-分析的作用和意義1.提供科學(xué)的證據(jù)循證醫(yī)學(xué)中的證據(jù)，按質(zhì)量和可靠程度可分為五級（可靠性依次降低）：一級：所有隨機(jī)對照試驗(yàn)的系統(tǒng)評價/Meta-分析。二級：單個的樣本量足夠的RCT結(jié)果。三級：設(shè)有對照組但未用隨機(jī)方法分組。四級：無對照的病例觀察。五級：專家意見。在治療方面，國際公認(rèn)大樣本RCT和RCT的系

4、統(tǒng)評價結(jié)果是證明某種療法的有效性和安全性最可靠的依據(jù)（金標(biāo)準(zhǔn)）。目前發(fā)達(dá)國家已將SR和RCT作為制定治療指南的主要依據(jù)。但在沒有這些金標(biāo)準(zhǔn)的情況下其他非隨機(jī)對照試驗(yàn)的臨床研究及其SR也可作為參考依據(jù)但可靠性降低。102. 增大樣本含量，減少各種偏倚和隨機(jī)誤差，提高統(tǒng)計(jì)檢驗(yàn)效能，加強(qiáng)主要結(jié)論的說服力，尤其是對樣本量較小的臨床實(shí)驗(yàn)研究。大規(guī)模RCT消耗人力, 財(cái)力和時間多數(shù)單位沒有條件作大規(guī)模RCT多數(shù)單個RCT樣本量小而不能得出準(zhǔn)確可靠的結(jié)論系統(tǒng)評價/Meta-分析聯(lián)合單個RCT，增大了樣本含量高質(zhì)量的Meta分析與高質(zhì)量的大樣本多中心臨床實(shí) 驗(yàn)一樣被認(rèn)為質(zhì)量最高的證據(jù)級別。系統(tǒng)評價/Meta-

5、分析的作用和意義112. 評價文獻(xiàn)中各研究結(jié)果的一致性，試圖解決或調(diào)和不同研究間的矛盾結(jié)論。3. 發(fā)現(xiàn)某些單個研究未闡明的問題。系統(tǒng)評價/Meta-分析的作用和意義12系統(tǒng)評價/Meta-分析的作用和意義5.使證據(jù)的使用更加方便信息爆炸時代：每年有200多萬篇有關(guān)生物醫(yī)學(xué)的文獻(xiàn)發(fā)表在2萬多種雜志上。臨床醫(yī)生需要系統(tǒng)評價/meta-分析有效地整合已有的信息，從而為合理的決策提供依據(jù)。特別是當(dāng)眾多試驗(yàn)結(jié)果相互矛盾時，系統(tǒng)評價/meta-分析可以提供一個可靠的綜合的答案。14Meta分析的意義1 病因研究2 干預(yù)措施的效果評價3 臨床診治的成本效益問題4 衛(wèi)生策略的效果評價15要求：簡單、扼要、明確

6、 一個好的問題應(yīng)包括四各關(guān)鍵因素：何種病人采取何種干預(yù)措施采用何種判斷療效的指標(biāo)研究設(shè)計(jì) 例如：急性缺血性腦卒中用溶栓療法是否能降低病死率和殘廢率？隨機(jī)對照試驗(yàn)的系統(tǒng)評價。1. 提出問題，明確研究目的172. 查找收集文獻(xiàn)采用多種檢索工具，盡可能收集所有相關(guān)文獻(xiàn)。應(yīng)事先制定一個收集文獻(xiàn)的策略，例如可從以下幾個方面收集臨床試驗(yàn)：計(jì)算機(jī)檢索:The Cochrane Library、Medline、Embase及中文醫(yī)學(xué)文獻(xiàn)數(shù)據(jù)庫等。人工檢索查閱學(xué)術(shù)會議論文集由國際、國內(nèi)的臨床試驗(yàn)資料庫提供資料由藥廠提供資料從臨床試驗(yàn)研究者或其他人員獲得信息。注意：除了收集已發(fā)表的文獻(xiàn)外，還應(yīng)當(dāng)收集未發(fā)表的資料

7、以避免“發(fā)表偏倚”。183. 確定納入和排除標(biāo)準(zhǔn)常用于確定納入標(biāo)準(zhǔn)的因素：研究設(shè)計(jì)類型：RCT，病例對照，隊(duì)列研究， 橫斷面資料病人種類干預(yù)措施判定療效的指標(biāo) 19 4. 資料的提取確定了納入的研究后，需要對各研究資料進(jìn)行提取。可設(shè)計(jì)一個統(tǒng)一的表格用于收集記錄各種信息及數(shù)據(jù)。包括文獻(xiàn)類型、發(fā)表年月、設(shè)計(jì)方法、樣本大小、干預(yù)措施、結(jié)果等。在資料提取的過程中應(yīng)注意兩個問題：1）有時在發(fā)表的文章中缺乏所需要的數(shù)據(jù)，這時需要與原作者聯(lián)系獲得；2）為了避免偏倚，確保質(zhì)量，應(yīng)由兩人獨(dú)立按選擇標(biāo)準(zhǔn)進(jìn)行資料提取，然后對提取的資料進(jìn)行比較核對。 205. 各研究的質(zhì)量評估如果將質(zhì)量不同的試驗(yàn)結(jié)果進(jìn)行合并分析，可

8、能會得出錯誤的結(jié)論。判斷研究 “質(zhì)量”的好壞的依據(jù)：該研究的設(shè)計(jì)和實(shí)施中可能防止產(chǎn)生系統(tǒng)誤差(偏倚)的程度。目前尚無統(tǒng)一的標(biāo)準(zhǔn)用于試驗(yàn)研究的質(zhì)量評估，以下四個方面的評估是最基本的：選擇偏倚的防止：隨機(jī)分配兩組干預(yù)措施以外的其他處理措施是否一致在下結(jié)論時是否包括了所有進(jìn)入試驗(yàn)的病人：退出和失訪測量偏倚的控制：盲法的實(shí)施216. 統(tǒng)計(jì)學(xué)處理系統(tǒng)評價與一般文獻(xiàn)綜述相區(qū)別的一個重要特點(diǎn)之一是：方法不是簡單地羅列、比較收集到的研究結(jié)果，而是在有條件的情況下把收集到的結(jié)果以數(shù)據(jù)的形式進(jìn)行統(tǒng)計(jì)學(xué)處理，即進(jìn)行Meta-分析。2224 不同資料的統(tǒng)計(jì)方法： 1.分類資料 固定效應(yīng)模型(fixed effect

9、model)：適用于同質(zhì)性的資料。 隨機(jī)效應(yīng)模型(random effect model)：適用于異質(zhì)性的資料。 2.計(jì)量資料的分析 合并檢驗(yàn)(combined tests) 效應(yīng)大小的測量效應(yīng)指數(shù)6. 統(tǒng)計(jì)學(xué)處理25 7. 敏感性分析 比較兩種不同方法對相同試驗(yàn)進(jìn)行的Meta分析是否會得出不同結(jié)果的過程，稱為敏感性分析。 目的是了解系統(tǒng)評價的結(jié)果是否可靠。 通過分組（如按研究類型、質(zhì)量、發(fā)表年份等進(jìn)行分組）進(jìn)行分析；也可對同一組資料采用不同的統(tǒng)計(jì)方法進(jìn)行分析以觀察結(jié)果的變化。 如敏感性分析未從實(shí)質(zhì)上改變結(jié)果，說明結(jié)果可信，否則下結(jié)論時應(yīng)謹(jǐn)慎。278.結(jié)果的綜合、解釋與討論結(jié)論的主要內(nèi)容包括：

10、說明根據(jù)所包括的研究是否能夠作出某一療法是否有效或某種因素是否為危險(xiǎn)因素的結(jié)論；是否可以在臨床實(shí)踐中推廣；如果現(xiàn)有資料尚不足以下結(jié)論，那么可看出什么趨勢？提出是否應(yīng)該進(jìn)一步研究的建議。 28系統(tǒng)評價/Meta分析中的偏倚 1.發(fā)表偏倚 2.文獻(xiàn)收集偏倚 3.納入標(biāo)準(zhǔn)偏倚 4.篩選者偏倚 5.數(shù)據(jù)摘錄偏倚29系統(tǒng)評價實(shí)例Advice on low-fat diets for obesity30How effective are low-fat diets in achieving sustained weight loss when used for the purpose of weight

11、loss in obese or overweight people？To assess the effects of advice on low-fat diets as a means of achieving sustained weight loss, using all available randomised clinical trials.1. 提出問題，明確研究目的31 2. 查找文獻(xiàn)Search strategy for identification of studies The following sources were included in the literatur

12、e search process: The Cochrane Controlled Trials Register (Cochrane Library Issue 2, 2001), MEDLINE, EMBASE, Science Citation Index, bibliographies, and handsearching. The original searches of MEDLINE and EMBASE were conducted from the beginning of each database until January 2001. The search was up

13、dated in February 2002. ELECTRONIC SEARCHES The Cochrane Library includes search results from several electronic databases and handsearch results from multiple journals. It was searched using the following strategy。 HANDSEARCHING323. 確定納入和排除標(biāo)準(zhǔn)The titles and abstracts of the records identified throug

14、h mentioned search strategies were independently screened by two investigators. Exclusive rules: (a) was not a report of a randomized controlled trial, (b) did not address a low fat diet, (c) did not have a follow-up period that was at least six months in duration, (d) was in children less than 18 y

15、ears old. 33When a title/abstract could not be rejected with certainty, the full text of the article was obtained for further evaluation. The full text of all selected articles was examined independently by two investigators to identify all relevant trials. Differences in opinion were resolved by co

16、nsensus.3. 確定納入和排除標(biāo)準(zhǔn)344. 資料的提取(DATA EXTRACTION)Three reviewers independently extracted data from the studies. Differences were resolved by the fourth reviewer. The data extraction form incorporated the following aspects of the trial: 1. General Information - full citation, language; 2. Trial charact

17、eristics - study design, study setting, duration, randomisation (and method), allocation (and method), blinding (patients, people administering treatment, outcome assessors), check of blinding;353. Intervention - characteristics of diets in treatment and comparison groups, description of nutrition a

18、dvice, inclusion of behavioural or psychological component, advice regarding of physical activity, use of weight-reducing medication, duration of intervention and duration of follow-up, replicability of intervention;4. Patients - sampling method, inclusion and exclusion criteria, total number and nu

19、mber in comparison groups, sex, age, and other relevant baseline characteristics (weight, BMI, % body fat, serum lipids, blood pressure), similarity of groups at baseline (including any co-morbidity), assessment of compliance, withdrawals/losses to follow-up (reasons/description), subgroups;4. 資料的提取

20、(DATA EXTRACTION)365. Outcomes - weight, BMI, % body fat, serum lipids (total cholesterol, LDL cholesterol, HDL cholesterol, VLDL cholesterol, triglycerides), glycemic profile (glycated haemoglobin, fasting plasma glucose), blood pressure, any other outcomes assessed;6. Results - for outcomes and ti

21、mes of assessment (including a measure of variation), if necessary converted to measures of effect specified below; intention-to-treat analysis.Authors were not contacted for further information.4. 資料的提取(DATA EXTRACTION)37All studies were categorised as C indicating that one or more criteria were no

22、t met.RANDOMIZATION: All studies were parallel design randomised controlled trials although the method of randomisation and allocation concealment processes were not described in any of the trials apart from McManus 2001. BLINDING: Blinding of caregivers and participants was either not done or not c

23、learly specified. In most studies it was unclear whether outcome assessment was conducted blindly and who actually measured the outcome factors.DROP-OUTS AND WITHDRAWALS: Drop-outs and withdrawals varied from 11% to 40%. 5. 各研究的質(zhì)量評估386. 統(tǒng)計(jì)學(xué)分析HETEROGENEITY：The results for the test of heterogeneity fo

24、r the overall results (low fat diet versus control diet in all participants) were significant for weight loss at each of the time periods and significant for change in BMI at the six and 12 month follow-up. Weighted Mean Differences (WMD) have been calculated using the random effects method.SUBGROUP

25、 ANALYSES：No subgroup analyses were conducted due to the small number of studies identified and the differences among the studies.3940417.敏感性分析SENSITIVITY ANALYSES：No sensitivity analyses were conducted due to the small number of studies identified and the generally poor quality of the studies. 428. 結(jié)果的綜合、解釋與討論SUMMARY OF FINDINGS：The results of this systematic review show that there is no real difference between a low fat diet a