Abstract:
|
Publication bias is a major threat to the validity of meta-analyses (MAs). Clinical studies with favorable results are more likely published and thus overestimate the synthesized results. The trim-and-fill (TF) method is a popular tool to detect and adjust for publication bias. Based on real-world MAs, we provide practical guidelines and recommendations for using the three estimators (R_0, L_0 and Q_0) to implement the TF method. A resampling method was proposed to calculate p-values for all three estimators. We applied the method to 29,932 MAs from the Cochrane Library, empirically evaluated its overall performance, and carefully explored potential issues. In many MAs, the significance of heterogeneity and overall effect sizes changed after adding imputed missing studies. L_0 and Q_0 failed to converge in a few MAs which contained studies with identical effect sizes. Also, outliers and pre-specified directions of missing studies could have influential impacts on the results. We recommend the TF method to be performed cautiously via MA software. Sensitivity analyses are encouraged to examine the effects of different estimators and outlying studies.
|