高級(jí)醫(yī)學(xué)統(tǒng)計(jì)學(xué)：樣本含量的估計(jì)

樣本含量的估計(jì)研究生《醫(yī)學(xué)統(tǒng)計(jì)學(xué)2》課程主要內(nèi)容最小樣本含量影響樣本含量估計(jì)的因素樣本含量的估計(jì)方法常用樣本含量的估計(jì)軟件介紹SASStataPASSR….最小樣本含量最小樣本含量指分辨“指標(biāo)A<>指標(biāo)B”所需的最小例數(shù)。確切的最小樣本含量理論上而言是不可知的。影響最小樣本含量的因素?cái)?shù)據(jù)種類；真實(shí)效應(yīng)的大?。褐笜?biāo)間真實(shí)差異；相關(guān)程度；變異度；研究質(zhì)量；設(shè)計(jì)方法；各組例數(shù)是否平衡；α和β最小樣本含量的計(jì)算參數(shù)估計(jì)時(shí)所需樣本量的計(jì)算病例對(duì)照設(shè)計(jì)所需樣本量的計(jì)算隊(duì)列設(shè)計(jì)所需樣本量的計(jì)算隨訪資料所需樣本量的計(jì)算參數(shù)估計(jì)的最小樣本含量計(jì)算以均數(shù)為例確定容許誤差：樣本均數(shù)和總體均數(shù)最大相差控制在什么范圍；一般取可信區(qū)間寬度的一半。確定控制容許誤差的概率αμ-uα/2uα/2α/2α/2均數(shù)例：某廠有6500人，擬用單純隨機(jī)抽樣了解該廠職工的白細(xì)胞計(jì)數(shù)的平均水平。希望絕對(duì)誤差不超過100/mm3。以往資料表明，職工白細(xì)胞標(biāo)準(zhǔn)差為950/mm3α取0.05。率的抽樣例：我國婦女現(xiàn)階段峰值年生育率在0.3上下波動(dòng)；容許誤差定為0.015；α取0.05。其他抽樣方式下樣本量估計(jì)專用公式分層抽樣整群抽樣設(shè)計(jì)效應(yīng)(deff)

假設(shè)檢驗(yàn)中的樣本含量估計(jì)：原理以單樣本的u檢驗(yàn)為例

0界值

/2

1-

0+d拒絕H0不拒絕H0兩樣本均數(shù)的比較的樣本含量估計(jì)：I類誤差，常取0.05

：II類誤差，常取0.20，0.101-：把握度：標(biāo)準(zhǔn)差，個(gè)體變異：臨床上能接受的最小差別（也可理解為本質(zhì)差異）221)(2ú?ùê?é+=dsbauunn＝兩個(gè)率比較的樣本含量估計(jì)：I類誤差，常取0.05

：II類誤差，常取0.20，0.101-：把握度病例-對(duì)照設(shè)計(jì)的樣本量估計(jì)：成組設(shè)計(jì)p1、p0為病例和對(duì)照組的暴露率；p=

(p1+p0)/2C為病例組和對(duì)照組樣本量的比例病例-對(duì)照設(shè)計(jì)的樣本量估計(jì)：配比設(shè)計(jì)p1、p0為病例和對(duì)照組的暴露率；1:m配比時(shí)病例-對(duì)照設(shè)計(jì)樣本量估計(jì)：實(shí)例吸煙與慢阻肺關(guān)系的病例-對(duì)照研究對(duì)照人群中吸煙率48%;RR=3；α=0.05，β=0.10p0=0.48；根據(jù)RR可以算出p1=0.7347，p=0.6073總共需要69名病例和69名對(duì)照如果按照1:1配比設(shè)計(jì)總共需要44對(duì)樣本。隊(duì)列設(shè)計(jì)的樣本含量估計(jì)隊(duì)列設(shè)計(jì)的比較思路實(shí)際上是兩個(gè)率的比較；故其樣本量的計(jì)算方法與率的假設(shè)檢驗(yàn)所需樣本量公式相同。隨訪研究中的樣本含量估計(jì)對(duì)于隨訪資料，除了前面需要的參數(shù)外，可能還需要中位生存時(shí)間HazardRatio入組時(shí)間長短入組模式……常用樣本含量估計(jì)軟件簡介StataSASPASSRPSSTATAStata的樣本量模塊Stata的sampsi命令SASSAS

PowerandSampleSizeSAS

PowerandSampleSizePowerAnalysisandSampleSize(PASS)PASSPASSTwoIndependentProportions(NullCase)PowerAnalysisPage/Date/Time 12012/10/2423:12:01NumericResultsofTestsBasedontheDifference:P1-P2H0:P1-P2=0.H1:P1-P2=D1<>0.TestStatistic:Ztestwithpooledvariance Sample Sample Prop|H1 Prop Size Size Grp1or Grp2or Diff Diff Grp1 Grp2 Trtmnt Control ifH0 ifH1 Target Actual Power N1 N2 P1 P2 D0 D1 Alpha Alpha Beta0.8128 39 39 0.8000 0.5000 0.0000 0.3000 0.0500 0.0536 0.1872Note:exactresultsbasedonthebinomialwereonlycalculatedwhenbothN1andN2werelessthan100.References…….ReportDefinitions'Power'istheprobabilityofrejectingafalsenullhypothesis.Itshouldbeclosetoone.'N1andN2'arethesizesofthesamplesdrawnfromthecorrespondingpopulations.'P1'istheproportionforgrouponeunderH1.Thisisthetreatmentorexperimentalgroup.'P2'istheproportionforgrouptwo.Thisisthestandard,reference,orcontrolgroup'TargetAlpha'istheprobabilityofrejectingatruenullhypothesisthatwasdesired.'ActualAlpha'isthevalueofalphathatisactuallyachieved.'Beta'istheprobabilityofacceptingafalsenullhypothesis.SummaryStatementsGroupsamplesizesof39ingrouponeand39ingrouptwoachieve81%powertodetectadifferencebetweenthegroupproportionsof0.3000.Theproportioningroupone(thetreatmentgroup)isassumedtobe0.5000underthenullhypothesisand0.8000underthealternativehypothesis.Theproportioningrouptwo(thecontrolgroup)is0.5000.Theteststatisticusedisthetwo-sidedZtestwithpooledvariance.Thesignificancelevelofthetestwastargetedat0.0500.Thesignificancelevelactuallyachievedbythisdesignis0.0536.PASS做PowerAnalysis TwoIndependentProportions(NullCase)PowerAnalysisPage/Date/Time 12012/10/2423:14:53NumericResultsofTestsBasedontheOddsRatio:O1/O2H0:O1/O2=1.H1:O1/O2=OR1<>1.TestStatistic:Ztestwithpooledvariance Sample Sample Prop|H1 Prop Size Size Grp1or Grp2or O.R. O.R. Grp1 Grp2 Trtmnt Control ifH0 ifH1 Target Actual Power N1 N2 P1 P2 OR0 OR1 Alpha Alpha Beta0.0638 50 50 0.5238 0.5000 1.000 1.100 0.0500 0.0569 0.93620.0699 100 100 0.5238 0.5000 1.000 1.100 0.0500 0.0560 0.93010.0696 150 150 0.5238 0.5000 1.000 1.100 0.0500 0.93040.0763 200 200 0.5238 0.5000 1.000 1.100 0.0500 0.92370.0830 250 250 0.5238 0.5000 1.000 1.100 0.0500 0.91700.0897 300 300 0.5238 0.5000 1.000 1.100 0.0500 0.91030.0965 350 350 0.5238 0.5000 1.000 1.100 0.0500 0.90350.1033 400 400 0.5238 0.5000 1.000 1.100 0.0500 0.89670.0824 50 50 0.5455 0.5000 1.000 1.200 0.0500 0.0569 0.91760.1073 100 100 0.5455 0.5000 1.000 1.200 0.0500 0.0560 0.89270.1234 150 150 0.5455 0.5000 1.000 1.200 0.0500 0.87660.1487 200 200 0.5455 0.5000 1.000 1.200 0.0500 0.85130.1742 250 250 0.5455 0.5000 1.000 1.200 0.0500 0.82580.1997 300 300 0.5455 0.5000 1.000 1.200 0.0500 0.8003FreeLunch!高帥富買SAS、PASS窮吊絲怎么辦？/Rpwr

packagefunctionpowercalculationsforpwr.2p.testtwoproportions(equaln)pwr.2p2n.testtwoproportions(unequaln)pwr.anova.testbalancedonewayANOVApwr.chisq.testchi-squaretestpwr.f2.testgenerallinearmodelpwr.p.testproportion(onesample)pwr.r.testcorrelationpwr.t.testt-tests(onesample,2sample,paired)pwr.t2n.testt-test(twosampleswithunequaln)PowerandSampleSize(PS)遺傳統(tǒng)計(jì)學(xué)中的樣本量估計(jì)(Quanto)Quanto特色可以估計(jì)各種設(shè)計(jì)下的樣本量；可針對(duì)于一般關(guān)聯(lián)性研究或交互作用研究；可設(shè)定不同遺傳模型；樣本含量計(jì)算中需要考慮的問題多重比較的校正復(fù)雜樣本量計(jì)算多組模擬能否在試