劑量反應(yīng)數(shù)據(jù)的Meta分析

劑量反應(yīng)數(shù)據(jù)的Meta分析第1頁/共17頁一、概述—乳糖攝入與卵巢癌2023/3/22第2頁/共17頁一、概述—文獻(xiàn)檢索2023/3/23第3頁/共17頁一、概述—信息提取2023/3/24第4頁/共17頁一、概述—信息提取idauthoryearstudyadjrrlbubdosecasen1Engle1991CC111015501Engle1991CC0.90.42.25.521561Engle1991CC1.30.62.911.535541Engle1991CC0.90.422216522Risch1994CC1110972322Risch1994CC1.040.711.5391072502Risch1994CC0.860.581.28161022432Risch1994CC1.070.721.59301432843Webb1998CC11101282923Webb1998CC1.010.711.43111332973Webb1998CC1.060.741.51161342963Webb1998CC1.40.982241773283Webb1998CC0.970.671.4146.51493174Goodman2002CC11101402924Goodman2002CC0.550.551.0961402924Goodman2002CC0.670.470.95121402924Goodman2002CC0.610.420.8923.81402925Salazar2002CC1110352435Salazar2002CC0.770.441.3728282415Salazar2002CC0.540.290.9949.5212296Cozen2002CC1110781836Cozen2002CC0.90.5791.3997.5571506Cozen2002CC0.940.6291.40615832126Cozen2002CC1.0010.6561.52329.6721732023/3/25第5頁/共17頁二、統(tǒng)計(jì)分析2023/3/26第6頁/共17頁二、統(tǒng)計(jì)分析—分析思路劑量反應(yīng)Meta分析的本質(zhì)就是回歸分析，其中有一個(gè)重要的假設(shè)就是通過選擇合適的鏈接函數(shù)，來定量評(píng)價(jià)效應(yīng)量與暴露劑量的關(guān)系如果非線性回歸能滿足要求，就可以進(jìn)行非線性劑量反應(yīng)的Meta分析，否則就為線性劑量反應(yīng)的Meta分析一般先非線性劑量反應(yīng)關(guān)系Meta分析，然后根據(jù)計(jì)算出來的關(guān)鍵變量對(duì)其線性情況進(jìn)行統(tǒng)計(jì)學(xué)檢驗(yàn)，根據(jù)檢驗(yàn)結(jié)果判斷效應(yīng)量與暴露劑量符合線性或非線性關(guān)系，確定進(jìn)行線性或非線性劑量反應(yīng)關(guān)系Meta分析2023/3/27第7頁/共17頁二、統(tǒng)計(jì)分析—分析思路定量合并的模型隨機(jī)效應(yīng)模型固定效應(yīng)模型STATA分析是基于廣義最小二乘法（generalizedleastsquaresmethod）即通過glst命令實(shí)現(xiàn)畫圖是xblc命令實(shí)現(xiàn)2023/3/28第8頁/共17頁二、統(tǒng)計(jì)分析—產(chǎn)生新變量2023/3/29gendoublelogrr=log(adjrr)gendoubleloglb=log(lb)gendoublelogub=log(ub)gendoublese=((logub-loglb)/(2*invnorm(0.975)))第9頁/共17頁二、統(tǒng)計(jì)分析—非線性劑量反應(yīng)關(guān)系_pctiledose,percentile(5356595)retlistmksplinedoses=dose,knots(07.51646.5)cubicdisplayknotsglstlogrrdoses*,se(se)cov(ncase)pfirst(idstudy)eform2023/3/210隨機(jī)效應(yīng)模型random第10頁/共17頁二、統(tǒng)計(jì)分析—非線性劑量反應(yīng)關(guān)系線性與非線性檢驗(yàn)STATA命令testparmdoses2doses32023/3/211第11頁/共17頁二、統(tǒng)計(jì)分析—非線性劑量反應(yīng)關(guān)系predictnllogrrwithref=_b[doses1]*doses1+_b[doses2]*doses2+_b[doses3]*doses3,ci(lb1ub1)genrrwithref=exp(logrrwithref)genlbwithref=exp(lb1)genubwithref=exp(ub1)levelsofdose,local(level)xblcdoses*,c(dose)at(05.567.591111.51215162223.79999923706055242829.600000381469733046.549.5)ref(0)eformtwoway(linelbwithrefubwithrefrrwithrefdose,sortlp(shortdashshortdash1)lc(blackblackblack))2023/3/212第12頁/共17頁二、統(tǒng)計(jì)分析—線性劑量反應(yīng)關(guān)系glstlogrrdose,se(se)cov(ncase)pfirst(idstudy)eform2023/3/213第13頁/共17頁二、統(tǒng)計(jì)分析—線性劑量反應(yīng)關(guān)系predictnllrr_lin=_b[dose]*dose,ci(lohi)genrr_lin=exp(lrr_lin)genlblin=exp(lo)genublin=exp(hi)levelsofdose,local(level)xblcdose,c(dose)at(05.567.591111.51215162223.79999923706055242829.600000381469733046.549.5)ref(0)eformtwoway(linelblinublinrr_lindose,sortlp(shortdashsh