Web version. 2.0 15 May 2006 StatsDirect ver. 2.0 15 May 2006 2 2 2 Meta-Analysis for Beginners by using the StatsDirect ver. 2.0 15 May 2006 Yukari KAMIJIMA 1), Ataru IGARASHI 2), Kiichiro TSUTANI 2) 1) Department of Pharmacoepidemiolog y, Faculty of Medicine, University of Tokyo 2) Department of Pharmacoeconomics, Graduate School of Pharmaceutical Sciences, University of Tokyo
Meta-Analysis: MA 2002 5,000 MA Systematic Review: SR 2) MA 3) MA 3 1) commercial Comprehensive Metaanalysis, DSAT, MetaWin, Metaxis 2) free available Easy MA, Meta, Meta-Analysist, Meta-Test, RevMan 3) MA Bugs and WinBUGS, SAS, S-Plus, Stata, StatXact, StatsDirect, True Epistat, StatsDirect Kaplan-Meire Chapter 1: StatsDirect Chapter 2: StatsDirect Chapter 3: StatsDirect StatsDirect 1) Multivariate meta-analysis 2) Cumulative meta-analysis 3) Baysian meta-analysis i
ii
StatsDirect Chapter 1: StatsDirect 1 Chapter 2: StatsDirect 13 Chapter 3: 24 34 iii
Chapter 1: StatsDirect (1) StatsDirect http://www.statsdirect.com/ (2) 10 Try 1-1 1
(3) Try 1) 2) [click here to read the terms and conditions] StatsDirect 3) StatsDirect Download 1-2 1) 2) 3) 2
(4) [click here] 1-3 3
(5) 1) 2) 3) Submit 1-4 1) 2) 4
(6) 2 2 3 2 3 1-5 5
(7) WSetupStatsDirect.EXE SetupStatsDirect.EXE 2 3 [Next] 1-6 6
(8) [ accept] 1-7 7
(9) Program Files 1-8 8
(10) StatsDirect 1-9 9
(11) [OK] StatsDirect 1-10 10
(12) 10 StatsDirect p.5 10 StatsDirect URL StatsDirect 3 2003 12 24 120 StatsDirect * 1 99 145 17,400 5 299 435 52,200 10 499 730 87,600 1 179 265 31,800 5 499 730 87,600 10 799 1,165 139,800 ** 49 1,165 8,640 * ** FAX VISA Master Amex JCB 11
URL: http://www.statsdirect.com/buy.htm 1-11 12
Chapter 2: StatsDirect StatsDirect StatsDirect 1) Odds Ratio 2) Peto Odds Ratio Peto 3) Relative Risk 4) Risk Difference 5) Summary 6) Effect Size 7) Incidence Rate 7 p.20 Odds Ratio 13
(1) StatsDirect 1) 2) [OK] 10 2-1 2) 1) 14
(2) [OK] 1) 10 [OK] (3) StatsDirect 2) 10 URL StatsDirect 2-2 2) 1) 15
(3) test.sdw 1) [New StatsDirect Work Book] 2) [OK] 2-3 aspirin 7 Table 1 Table1 Table 2-1 7 16
Table 1 Trial aspirin group placebo group event event total event event total MRC-1 49 566 615 67 557 624 CDP 44 714 758 64 707 771 MRC-2 102 730 832 126 724 850 GASP 32 285 317 38 271 309 PARIS 85 725 810 52 354 406 AMIS 246 2021 2267 219 2038 2257 ISIS-2 1570 7017 8587 1720 6880 8600 Table 2-1 Trial MRC-1 event event total aspirin 49 566 615 placebo 67 557 624 1229 Table 2-2 Trial CDP event event total aspirin 44 714 758 placebo 64 707 771 1529 Table 2-3 Trial MRC-2 event event total aspirin 102 730 832 placebo 126 724 850 1682 Table 2-4 Trial GASP event event total aspirin 32 285 317 placebo 38 217 309 626 Table 2-5 Trial PARIS event event total aspirin 85 725 810 placebo 52 354 406 1216 Table 2-6 Trial AMIS event event total aspirin 246 2021 2267 placebo 219 2038 2257 4524 Table 2-7 Trial ISIS-2 event event total aspirin 1570 7017 8587 placebo 1720 6880 8600 17187 17
Table 1 StatsDirect StatsDirect Stratum Label Total number of patients in EXPERIMENTAL group Number of patients with event in EXPERIMENTAL group event Total number of patients in CONTROL group Number of patients with event in CONTROL group event 5 5 Table 3 Table 3 Trial aspirin group placebo group total event total event MRC-1 615 49 624 67 CDP 758 44 771 64 MRC-2 832 102 850 126 GASP 317 32 309 38 PARIS 810 85 406 52 AMIS 2267 246 2257 219 ISIS-2 8587 1570 8600 1720 18
(4) StatsDirect Excel Excel Table 3 A Trial B Total number of patients in aspirin group aspirin C Number of patients with event in aspirin group aspirin event D Total number of patients in placebo group placebo E Number of patients with event in placebo group placebo event 2-4 19
(5) 1) [Analysis] 2) [Meta-Analysis] 3) [Odds Ratio] 2-5 1) 3) 2) 20
(6) 1) Select TOTAL numbers of subjects in EXPERIMENTAL groups [1 column] 2) B B StatsDirect 3) [OK] [OK] / Select numbers of EXPERIMENTAL subjects with DISEASE/OUTCOME [1 column] C [OK] 4) 5) Select stratum LABELS [cancel for default [1 column]] A [OK] 2-6 1) 3) 2) 21
(7) [ ] 2-7 22
(8) bias assessment plot vertical axis CI confidence interval standard error [standard error] [include CI interval if relevant] [OK] 2-8 (9) 23
Chapter 3: [ ] 4) 4. 2002 1) Table 3 p.17 Table 3 Stratum 2 aspirin group event 3 placebo group event 4 aspirin group event 5 placebo group event 2) Odds ratio 95%CI 95% M-H Weight Mantel-Haenszel weight [p.72] Fixed effects [p.49] Random effects [p.49] 3) 4) 5) Fixed effects 6) Random effects Fixed effects model 5) Fixed effects Fixed effects model Random effects model 24
3-1 1) 2) 3) 4) 5) 25
StatsDirect Fixed effects Mantel-Haenszel Random effects DerSimonian-Laird [p.71] 3) Fixed effects Mantel-Haenszel chi-square Mantel-Haenszel Mantel-Haenszel pooled estimate of odds ratio Approximate 95%CI 95% StatsDirect Mantel-Haenszel 95% Robins, Breslow and Greenland method [p.159] 4) Fixed effects Conditional maximum likelihood estimate of pooled odds ratio Exact Fisher 95% confidence interval Fisher 95% Exact Fisher one sided P two sided P p 95% 4 5 Exact mid-p 95% confidence interval 95% Exact Fisher one sided P two sided P p 5) Non-combinability of odds ratios StatsDirect Breslow-Day Woolf p df degree of freedom k 26
3-1 1) 2) 3) 4) 5) 27
6) Random effects DerSimonian-Laird pooled odds ratio Approximate 95% CI 95% DerSimonian-Laird chi-square [p.72] 7) StatsDirect From regression of normalized effect vs. precision Egger From Kendall's test on standardized effect vs. variance Kendall Begg Mazumdar [p.121] Egger funnel plot α β 0 StatsDirect Intercept y approximate 95% CI 95% p funnel plot p.30 Begg Mazumdar Kendall Kendall tau StatsDirect tau Kendall p not robust, small sample 10% 28
3-1 3) 4) 5) 6) 7) 29
Fig.1 funnel plot funnel plot p.51 StatsDirect p.22 Part 2. (8) 7) Egger Egger Part 2. 8 [p.58 121] Fig.2 L'Abbe plot 5) aspirin group placebo group event L'Abbe plot 30
3-2 Fig.1 Fig.2 31
Fig.3 Fixed effects Fig.4 Random effects 32
3-3 Fig.3 Fig.4 33
34
StatsDirect p.13 20 StatsDirect 1) Odds Ratio Mantel-Haenszel DerSimonian-Laird 2) Peto Odds Ratio Peto 3) Relative Risk Mantel-Haenszel DerSimonian-Laird 4) Risk Difference Mantel-Haenszel DerSimonian-Laird 5) Summary 6) Effect Size 7) Incidence Rate 1) Odds Ratio Mantel-Haenszel DerSimonian-Laird StatsDirect 2) 7) p.20 3) StatsDirect Help kamijima-tky@umin.ac.jp 35
1). systematic review. 2003; 34(4): 210-6 2) Sterne JAC, Egger M, Sutton AJ, Meta-analysis software. In: Systematic reviews in health care: meta-analysis in context. BMJ Books, 2002 p.236-46 3).. 2003; 19(8) 4). 4. 2002. 5) Song F. Exploring heterogeneity in meta-analysis: Is the L Abbe plot useful? J Clin Epidemiology; 52(8): 725-30, 1999. 15 36
kamijima-tky@umin.ac.jp