ABSTRACT
86There are at least three sources of variation to consider before combining summary statistics across studies. They are inter-study variation, sampling error among studies, and study-level characteristics. Once the data have been assembled, simple inspection of the Forest plot is informative. Heterogeneity between study results should not be seen as purely a problem for systematic reviews, since it also provides an opportunity for examining why treatment effects differ in different circumstances. The sample size method used for meta-analysis employs a weighted average of the results in which the larger study generally has more influence than the smaller ones. Selection of a meta-analysis method for a particular analysis depends on the type of primary studies, choice of summary statistics, observed heterogeneity, the known limitations of the computational methods, and fixed effects versus random effects model. Fixed effects model is centered on making inferences for every population that have been sampled, then the outcomes are considered fixed and the only source of uncertainty is that resulting from the sampling of people into the studies. Pooling of study results under sample size method is mainly done under the assumption that, k samples are from a normal population. The inverse-variance method is used to pool binary, continuous, and correlation data. This approach has wide applicability since it can be used to combine any estimate that has standard error available. Mantel-–Haenszel methods have been shown to be more robust when data are sparse, and may therefore be preferable to the inverse-variance method. In combining odds ratio, an alternative to the Mantel-–Haenszel method is Peto.
