查看文章

Meta-analysis involves representing the results of each study by a numerical index of effect size and statistically combining these estimates across studies. Effect sizes used in meta-analysis include risk differences, risk ratios, odds ratios, standardized mean differences, and Pearson correlation coefficients. Computation of each of these effect sizes and their sampling variances is described and both fixed and random effects weighted means to combine estimates across studies are illustrated. The Mantel-Haenszel combination procedure is described as an alternative fixed effects method when studies have binary outcomes and small sample sizes. More complex combination procedures such as meta-regression are mentioned as are methods to detect and adjust for publication bias. Methods are illustrated with data from a published meta-analysis.