Assessment of heterogeneity and publication bias Heterogeneity

Assessment of heterogeneity and publication bias
Heterogeneity, the visible diversity in mean difference across studies was quantified using the I statistic (Higgins and Thompson 2002; Higgins et al. 2003). I2 lies between 0 and 100%, and I2 with negative values were assigned a value of zero (Higgins et al., 2003). An I2 with 25%, 50% and 75% were considered as small, moderate and large heterogeneity, respectively. We use random-effect model when the test for heterogeneity is significant, otherwise, estimated results based on a fixed-effect model were presented. In the current study, we assessed the presence of publication bias using Fail-Safe N and funnel plots. Funnel plot is a simple scatter plot of the intervention effect estimates from individual studies (horizontal axis) plotted against study precision (vertical axis; Sterne and Harbord, 2004). For small studies, the effect estimates will tend to scatter more widely at the bottom of the graph while narrows for larger studies (Sterne and Harbord, 2004). In the absence of publication bias, the funnel plot should resemble a symmetrical (inverted) funnel, but if there is presence of publication bias, for instance because of unpublished smaller studies without statistically significant effects, this will lead to an asymmetrical appearance of the funnel plot and a gap will be evident in a bottom corner of the graph (Higgins and Green, 2011). In this situation, the effect calculated in a meta-analysis will tend to overestimate the intervention effect (Sterne and Harbord, 2004). The more pronounced the asymmetry, the more likely it is that the bias will be substantial. Individual studies with large or aberrant results were identified from the forest and funnel plots and evaluated for factors that may have influenced the results.