Meta-analysis

1、Stata在Meta分析中的應(yīng)用席波席波山東大學(xué)公共衛(wèi)生學(xué)院山東大學(xué)公共衛(wèi)生學(xué)院系統(tǒng)綜述與Meta分析的關(guān)系系統(tǒng)綜述（systematic review）是一種按照既定納入標(biāo)準(zhǔn)廣泛收集某醫(yī)療衛(wèi)生問(wèn)題的相關(guān)研究，而后嚴(yán)格評(píng)價(jià)其質(zhì)量，并進(jìn)行定量合并分析或定性分析，得出綜合結(jié)論的研究方法。Meta分析（meta-analysis）是用于比較和綜合針對(duì)同一科學(xué)問(wèn)題研究結(jié)果的統(tǒng)計(jì)學(xué)方法，其結(jié)論是否有意義取決于納入研究的質(zhì)量，常用于系統(tǒng)綜述中的定量合并分析。為什么進(jìn)行為什么進(jìn)行Meta分析？分析？解決結(jié)果矛盾增加樣本量Meta分析步驟分析步驟選題-Novelty檢索文獻(xiàn)- (Pubmed and Emba

2、se (or Scopus)根據(jù)入選標(biāo)準(zhǔn)選擇合格的研究復(fù)習(xí)每個(gè)研究并進(jìn)行質(zhì)量評(píng)估- (如NOS、Jadad評(píng)分)提取信息，填寫(xiě)摘錄表，建立數(shù)據(jù)庫(kù)繪制森林圖-(Stata or Revman)異質(zhì)性檢驗(yàn)(如果有異質(zhì)性，采用Meta回歸或亞組分析探討來(lái)源)亞組分析敏感性分析檢驗(yàn)發(fā)表偏倚非非RCT研究質(zhì)量評(píng)價(jià)研究質(zhì)量評(píng)價(jià)The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomized studies in meta-analyses.http:/www.ohri.ca/programs/clinical_epide

3、miology/oxford.aspRCT研究質(zhì)量評(píng)價(jià)研究質(zhì)量評(píng)價(jià)Jadad量表，又稱(chēng)為Jadad評(píng)分或牛津評(píng)分系統(tǒng)，是獨(dú)立評(píng)價(jià)臨床試驗(yàn)方法學(xué)質(zhì)量的工具。在類(lèi)似的評(píng)價(jià)量表中，Jadad量表的使用最為廣泛。Jadad量表由 Alejandro Jadad-Bechara 制定，作為哥倫比亞的一名醫(yī)生，他還是牛津大學(xué)內(nèi)納菲爾德麻醉劑部研究減輕疼痛的研究員。Jadad和他的組員在1996年的臨床對(duì)照試驗(yàn)雜志上發(fā)表了一篇有關(guān)盲法效應(yīng)的文章。在該文章的附錄中，通過(guò)評(píng)價(jià)，給不同臨床試驗(yàn)評(píng)分，從最差的0分到最高的5分。Jadad認(rèn)為隨機(jī)對(duì)照試驗(yàn)是現(xiàn)代醫(yī)學(xué)研究的一大進(jìn)步，在一本2007年寫(xiě)的一本書(shū)中，他說(shuō)“這

4、是一種最簡(jiǎn)單，但又最有效、最具有革命性的研究形式”。修改后的修改后的Jadad量表（量表（1-3分視為低質(zhì)量，分視為低質(zhì)量，4-7分視為高質(zhì)量）分視為高質(zhì)量）例1 為了探討用Aspirin預(yù)防心肌梗塞(MI)后死亡的發(fā)生，美國(guó)在1976-1988年間進(jìn)行了7個(gè)關(guān)于Aspirin預(yù)防MI后死亡的研究，其結(jié)果見(jiàn)表格，其中6項(xiàng)研究的結(jié)果表明Aspirin組與安慰劑組的MI后死亡率的差別無(wú)統(tǒng)計(jì)學(xué)意義，僅1項(xiàng)結(jié)果表明Aspirin預(yù)防 MI后死亡有效。StudyPublication yearAspirin groupPlacebo groupTotalDeathsTotalDeathsMRC-1197

5、46154962467CDP19767584477164MRC-21979832102850126GASP19793173230938PARIS19808108540652AMIS198022672462257219ISIS-219888587157086001720Forest plot命令: metan death1 live1 death2 live2, or label (namevar=study, yearvar=year) metan death1 live1 death2 live2, or label (namevar=study, yearvar=year) xlab(0.

6、45,1,2.19) metan death1 live1 death2 live2, or label (namevar=study, yearvar=year) xlab(0.45,1,2.19) textsize(150)Overall (I-squared = 39.7%, p = 0.127)MRC-1 (1974)GASP (1979)IDISIS-2 (1988)CDP (1976)MRC-2 (1979)AMIS (1980)PARIS (1980)Study0.90 (0.84, 0.96)0.72 (0.49, 1.06)0.80 (0.49, 1.32)OR (95% C

7、I)0.89 (0.83, 0.97)0.68 (0.46, 1.01)0.80 (0.61, 1.06)1.13 (0.93, 1.37)0.80 (0.55, 1.15)100.003.181.80Weight72.883.105.6810.153.22%0.90 (0.84, 0.96)0.72 (0.49, 1.06)0.80 (0.49, 1.32)OR (95% CI)0.89 (0.83, 0.97)0.68 (0.46, 1.01)0.80 (0.61, 1.06)1.13 (0.93, 1.37)0.80 (0.55, 1.15)100.003.181.80Weight72.

8、883.105.6810.153.22% 1.4512.19Sensitivity analysis命令： generate logor=log(_ES) generate selogor=_selogES metainf logor selogor, eform id(study) random print 0.80 0.90 0.84 0.96 1.02 MRC-1 CDP MRC-2 GASP PARIS AMIS ISIS-2 Study ommited Meta-analysis fixed-effects estimates (exponential form) Publicati

9、on bias命令： metabias logor selogor, graph(begg)Beggs funnel plot with pseudo 95% confidence limits logors.e. of: logor-.50.5例2 Gotzsche收集了有關(guān)短程小劑量強(qiáng)的松 VS安慰劑或非甾體抗炎藥治療類(lèi)風(fēng)濕性關(guān)節(jié)炎的7個(gè)臨床隨機(jī)對(duì)照試驗(yàn)(RCTs)，觀察類(lèi)風(fēng)濕性關(guān)節(jié)炎患者的關(guān)節(jié)壓痛指數(shù)(rechies index)。First author Publication yearExperimental treatmentControl treatmentNoMe

10、anSDNoMeanSDJasni1968916.28.7938.112.8Dick19702417.682440.713Lee19732130.516.52141.419.8Berry19741213111223.711.1Lee19741814.612.41826.415.1Stenberg1992216.31.72111.12.5Geital19952010.84.72016.37.7Forest plot 命令： metan n1 mean1 sd1 n2 mean2 sd2, label (namevar=study, yearvar=year) random textsize(18

11、0)NOTE: Weights are from random effects analysisOverall (I-squared = 71.7%, p = 0.002)Jasni (1968)Berry (1974)IDGeital (1995)Dick (1970)Lee (1973)Stenberg (1992)Lee (1974)Study-1.34 (-1.87, -0.80)-2.00 (-3.16, -0.85)-0.97 (-1.82, -0.12)SMD (95% CI)-0.86 (-1.51, -0.21)-2.14 (-2.86, -1.43)-0.60 (-1.22

12、, 0.02)-2.25 (-3.03, -1.47)-0.85 (-1.54, -0.17)100.0010.4413.45Weight15.6714.9316.0114.2215.28%-1.34 (-1.87, -0.80)-2.00 (-3.16, -0.85)-0.97 (-1.82, -0.12)SMD (95% CI)-0.86 (-1.51, -0.21)-2.14 (-2.86, -1.43)-0.60 (-1.22, 0.02)-2.25 (-3.03, -1.47)-0.85 (-1.54, -0.17)100.0010.4413.45Weight15.6714.9316

13、.0114.2215.28% 0-3.1603.16Sensitivity analysis命令： metainf _ES _seES, eform id(study) random print 0.13 0.26 0.15 0.45 0.51 Jasni Dick Lee Berry Lee Stenberg Geital Study ommited Meta-analysis random-effects estimates (exponential form) Publication bias命令： metabias _ES _seES, graph(begg)Beggs funnel

14、plot with pseudo 95% confidence limits SMDs.e. of: SMD-3-2-10Source of heterogeneity命令： metareg _ES year n1 n2, wsse(_seES) metan n1 mean1 sd1 n2 mean2 sd2,label (namevar=study, yearvar=year) random textsize(160) by(group_n1)NOTE: Weights are from random effects analysis.Overall (I-squared =

15、71.7%, p = 0.002)Stenberg (1992)Berry (1974)Subtotal (I-squared = 31.6%, p = 0.232)Geital (1995)=20Lee (1974)Dick (1970)IDSubtotal (I-squared = 82.9%, p = 0.001)Lee (1973)Study20Jasni (1968)-1.34 (-1.87, -0.80)-2.25 (-3.03, -1.47)-0.97 (-1.82, -0.12)-1.14 (-1.74, -0.54)-0.86 (-1.51, -0.21)-0.85 (-1.

16、54, -0.17)-2.14 (-2.86, -1.43)SMD (95% CI)-1.44 (-2.27, -0.61)-0.60 (-1.22, 0.02)-2.00 (-3.16, -0.85)100.0014.2213.4539.1715.6715.2814.93Weight60.8316.01%10.44-1.34 (-1.87, -0.80)-2.25 (-3.03, -1.47)-0.97 (-1.82, -0.12)-1.14 (-1.74, -0.54)-0.86 (-1.51, -0.21)-0.85 (-1.54, -0.17)-2.14 (-2.86, -1.43)S

17、MD (95% CI)-1.44 (-2.27, -0.61)-0.60 (-1.22, 0.02)-2.00 (-3.16, -0.85)100.0014.2213.4539.1715.6715.2814.93Weight60.8316.01%10.44 0-3.1603.16例3 探討MC4R基因多態(tài)性與2型糖尿病的關(guān)系Frayling TM, Timpson NJ, Weedon MN, et al. A common variant in the FTO gene is associated with body mass index and predisposes to childho

18、od and adult obesity. Science 2007; 316:889894 FTO 肥胖 2型糖尿病 Adjusted for BMIOR=1.03, 95%CI 0.96-1.10, p=0.44OR=1.15, 95%CI 1.09-1.23, p=910-6OR=1.32, 95%CI 1.26-1.39, p=310-26Li H, Kilpelinen TO, Liu C,et al. Association of genetic variation in FTO with risk of obesity and type 2 diabetes with data from 96,551 East and South Asians. Diabetologia. 2012 Apr;55(4):981-95. gene lnor=ln(or) gene lnci1=ln(ci1) gene lnci2=ln(ci2) m