總結證據結果 & Meta-Analysis 102-1-29 1

Meta-analysis 上機操作 (102-2-19):電腦教室 Lecture (102-1-29) Meta-analysis 上機操作 (102-2-19):電腦教室 2

Summary of the relevant references 研究品質製表 品質項目於各欄，文獻則置於各列 (依出版年排序) 作者 年度 隨機分派 盲法 退出研究者之描述 族群完整的診斷檢查步驟 結果一年之後續追蹤 品質排序 序列 產生法 隱匿 Roonberg 1980 未陳述 不清楚 適當 不適當 3 Abel 1982 4 Wang 1983 6 Torok 1985 5 Micic 9 Alinmelk 1987 8 Sokol 1988 1 WHO 1992 2 Karuse 7 3

Meta-Analysis Why meta-analysis Heterogeneity of the studies Fixed effect methods for combining study estimate Random effect methods for combining study estimate Sensitivity analysis Presenting the results of a meta-analysis 4

Meta-analysis 積少成多 (n大是王道) 減少 sampling errors Homogenicity 檢測 一致:IPD (individual patient data), summary data結果一致 存在 heterogenicity: summary data (meta-analysis) 5

在meta-analysis中, study的 重要性 (貢獻性)依 weight衡量. Weight= 1/variance 在meta-analysis中, study的 重要性 (貢獻性)依 weight衡量. 6

9 126/139 -0.85 12.49 0.02 -1.1 to -0.59 study n d s Var (d) 95% CI 1 17/20 0.28 10.59 0.11 -0.37 to 0.93 2 39/40 -0.81 16.18 0.05 -1.27 to -0.35 3 15/15 -0.9 8.86 0.15 -1.65 to -0.15 . 9 126/139 -0.85 12.49 0.02 -1.1 to -0.59 7

Due to heterogenicity, we do Meta-Analysis Heterogenicity test: Chi-square test: p<0.05 異質性過高 個別分析,評估異質性來源 Chi-square test: p>0.05 進入 meta-analysis 異質程度: I2=(Q-df)/Q x 100% (df=study numbmer -1) low: <=0.2; middle: 0.2-0.5; high: >0.5 但書: 10篇studies以下,改 用 0.1當 cutpoint 8

P-value for heterogenicity >=0.5 Fixed-effect model 0.05-0.5 Random-effect model <0.05 評估異質性來源 9

Relative weight of study: 在random-effect model中, 小型study權重相對提高. Fixed-effect model: the between groups variance is ignored; only with-in study variance is considered Random-effect model: consider both the between- and with-study variation; 較保守策略 Relative weight of study: 在random-effect model中, 小型study權重相對提高. 10

11

Example 12

13

Forest plot 14

<0.05評估異質性 來源 15

16

17

P=0.065 random effect 18

Subgroup analysis 19

Male 20

Female 21

22

Publication bias: Funnel plot 對稱漏斗圖 較小型研究中的效果(空心圈)，其散射 (scatter)比較大型研究的(寬心圈)寬廣； 它們在較大型研究中觀察到之效果的兩 邊都會出現。效果在一個漏斗型之內。 不對稱漏斗圖 效果比在較大型研究中觀察到的(實心 圈)小(空心圈)的較小型研究，似乎不見 了。漏斗是縮短的。 ?研究不見了 23

Funnel Plot Precision by Log odds ratio 24

Funnel Plot Precision by Log odds ratio 25

Egger’s test Egger et al. (1997) proposed a test for asymmetry of the funnel plot. This is a test for the Y intercept = 0 from a linear regression of normalized effect estimate (estimate divided by its standard error) against precision (reciprocal of the standard error of the estimate). StatsDirect provides this bias indicator method with all meta-analyses. Please note that the power of this method to detect bias will be low with small numbers of studies.  檢測 intercept是否為”0” 26

delete some study Sensitivity analysis Sensitivity of results of inclusion of criteria: delete some study Sensitivity of results to meta-analysis methods (different weighting) Weighting method 27

28 Log scale

其他8個studies 9 個 studies 29

Sensitivity test 30

Summary Abstract Background Hypothesis Review methods Details of studies included Details of studies excluded Results of the review Analysis of the robustness of the results: sensitivity test Discussion: included publication bias 31

Thanks for your attention! 32