Meta分析中发表偏倚的Begg s检验、Egger s检验 及Macaskill …

1 Chin J Evid-based Med 2009, 9(8): 910-916 910 CJEBM 2009 Editorial Board of Chin J Evid-based Begg s Egger s Macaskill s SAS 1, 2 3 1,*1. 430030 2. 330006 3. 100050 Meta Meta Begg s Egger s Macaskill s Meta SAS Stata Meta Begg s Egger s Macaskill s Applying the SAS Program to the Complete Begg s Test, the Egger sTest, and the Macaskill s Test for Publication Bias of Meta-analysisZHENG Hui-lie1,2, WANG Zhong-xu3, WANG Zeng-zhen1,* 1.

2 Department of Epidemiology and Statistics, School of Public Health, Tongji Medical College of Huazhong University of Science and Technology, Wuhan 430030, China2. Department of Epidemiology and Statistics, School of Public Health, Medical College of Nanchang University, Nanchang 330006, China3. National Institute for Occupational Health and Poison Control, Chinese Center for Disease Control and Prevention, Beijing 100050, ChinaAbstract The conclusions of meta-analyses are susceptible to various of biases, and publication bias is one of such main bias.

3 Therefore, Checking for evidence of publication bias should be undertaken routinely at the preliminary stage of a meta-analysis. Begg s test, Egger s test, and Macaskill s test are usually used to objectively identify publication bias in meta-analyses. In order to conveniently use these methods, the SAS program of these three tests was designed in this paper. In order test practical data, the fact that the output of this program of SAS software was consisted with the output of STATA software was validated.

4 So, this program is an alternative way to do such hypothesis tests to identify the publication bias in words Meta-analysis; Publication bias; Begg s test; Egger s test; Macaskill s testMeta Meta Meta Meta publication bias [1-3] Meta Meta 973 2002CB512910 1976 Email: * Email.

5 [4] Meta funnel plot [1,5] Meta RevMan SAS SPSS Stata Meta [6] Begg s [7] Egger s [8] Macaskill s [9] Meta Stata Begg s Egger s Meta RevMan SAS SPSS SAS Meta 2009, 9(8): 910 916 911 2009 Egger s Begg s Macaskill s SAS 1 Begg s [7] Egger s [8] Macaskill s [9]Begg s Egger s Macaskill s Meta Begg s y* v y* v Kendall s [10] y* y*= (yi y)/vi*1/2 y= ( vi 1yi)/ vi 1 vi*= vi ( vj 1)

6 1 y v i j Meta Kendall s 1 +1 0 0 0 Begg s Kendall s y* v Kendall s Kendall s Kendall s Kendall s k(k 1)/2 k Meta SAS Kendall s Kendall s [10] (k(k 1)(2k+5)/18)1/2 Egger s ln(OR) var(ln(OR)) SMD/ var(SMD) y x 0 0 y / x 1/ Macaskill s y n Macaskill s 0 Meta 2 SAS Meta Egger s Begg s Macaskill s Shi [11]

7 OR Meta SAS 35 CYP1A1 GSTM1 20 GSTM1 GSTM1 1 Meta 1 Egger s P= Egger s P= SAS Begg s P= Macaskill s P= Meta Meta SAS Begg s Egger s Stata 1 GSTM1 [11] GSTM GSTM1 + GSTM GSTM1 + Gao Jianrui199827192545Hu Yiling199834252930 Chen Senqing199939294263 Lan Q200082406062 Chen SQ200156503967 Ying Chan200243136534 Wang Jingwen200397679091 Chang-Yeung M200412510411780 Luo Chenling200445182423 Liang Geyu200482707973Li Weiying20041279095105Gu Yanfei200410179102122 Yang XR2004108787564Li Dairong200557422739 Zeng Mian200556355140 Wang Na200545324562Ye Weiyun200523353329Li Ying200559396177 Qian Biyun200669395355 Wang Qiming200640161923 1 3159201 1311 184 Chin

8 J Evid-based Med 2009, 9(8): 910-916 912 CJEBM 2009 Editorial Board of Chin J Evid-based SAS /** **//* case_n1: case_n0: con_n1: con_n0: study_size: */option linesize=120;%let study_size=20;data eff_sizeset;input case_n1 case_n0 con_n1 con_n0@@;k=_n_;/*ln(OR) , , */sample_size=case_n0+case_n1+con_n0+con _n1;or=(case_n1*con_n0/(case_n0*con_n1)) ;eff_size=log(or);var_eff_size=1/case_n1 +1/case_n0+1/con_n0+1/con_n1;se_eff_size =sqrt(var_eff_size);/* egger s y x*/egger_y=eff_size/se_eff_size;egger_x= 1/se_eff_size;/* (ln(OR)) */inv_v=1/var_eff_size;eff_size_inv_v=ef f_size*inv_v;cards;27 19 25 45 34 25 29 3039 29 42 63 82 40 60 6256 50 39 67 43 13 65 3497 67 90 91 125 104 117 8045 18 24 23 82 70 79 73127 90 95 105 101 79 102 122108 78 75 64 57 42 27 3956 35 51 40 45 32 45 6223 35 33 29 59 39 61 7769 39 53 55 40 16 19 23;run;/** **//** **//* egger s */ods listing close.

9 Ods output parameterestimates=egg1(keep=Variable Estimate StdErr tValue Probt rename=(variable=STD_EFF Estimate=coefficient));proc reg;model egger_y=egger_x;run;ods listing;data eggset;set egg1;if _n_=1 then STD_EFF='bias';else STD_EFF='slope';CIL_95=coefficient-tinv( ,&study_size-2)*stderr;CIU_95=coefficien t+tinv( ,&study_size-2)*stderr;run;data egg2(where=(STD_EFF="slope")) egg3(where=(STD_EFF="bias"));set eggset; 1 GSTM1 Meta 2009, 9(8): 910 916 913 2009 run;/* egger s eggset*/data eggset;m e r g e e g g 2 e g g 3 ( r e n a m e = ( S T D _ E F F = S T D _ E F F 1 c o e f f i c i e n t = c o e f f i c i e n t 1 S t d E r r = S t d E r r 1 P r o b t = P r o b t 1 tValue=tValue1 CIL_95=CIL_951 CIU_95=CIU_951)); run;quit;/*begg's */ /* ln(OR) (sum_inv_v)*/data beg1;set eff_sizeset;sum0+inv_v;if _n_=&study_size then sum=sum0;else sum=0;run.

10 Proc sort; by descending k;run;data beg2;set beg1;sum_inv_v+sum;run;/* ln(OR) (eff_size_sum_ inv_v)*/data beg3;set beg2;sum1+eff_size_inv_v;if _n_=&study_size then sum=sum1;else sum=0;run;proc sort; by k;run;data beg4;drop sum sum0 sum1;set beg3;sum_eff_size_inv_v+sum;run;/* ln(OR) ln(OR)- v* y* */data eff_sizeset1;set beg4;y_mean=sum_eff_size_inv_v/sum_inv_v ;v_asterisk=var_eff_size-1/sum_inv_v;y_a sterisk=(eff_size-y_mean)/sqrt(v_asteris k);run;/* kendall's */proc corr data=eff_sizeset1 kendall outk=kenda(keep=var_eff_size) noprint;var y_asterisk var_eff_size; run;proc transpose data=kenda out=kendall3(keep=col3 col4 rename=(col3=n col4=tau));/* kendall's Begg s z P */data kendall;set kendall3;kend_score=n*(n-1)*tau/2;se_ken d_score=sqrt(n*(n-1)*(2*))