An Improved Method for Minimizing Impact of Publication Bias and Study Heterogeneity in Meta-Analysis

Nova Southeastern University, Davie, Florida, USA

Objective. To develop a novel method to identify and mitigate publication bias (PB) and study heterogeneity (SH) in meta-analysis (MA). Background. MA is the preferred method to combine and analyze results of multiple studies to achieve an estimate of effect size. MA is subject to PB and SH which can negatively impact effect size. Methods to address PB and SH include graphical examination of funnel plots for symmetry and sensitivity analyses. Methods. A funnel plot is a plot of intervention effect of individual studies plotted against variance effect size. We modified traditional funnel plot analysis with a procedure from regression called reverse-stepping covariate selection. MA is conducted with all selected studies, the overall odds ratio (OR) estimate and the individual study ORs are then plotted. The plot is inspected, and the individual study whose confidence interval (CI) does not overlap the overall OR and CI are eligible for removal. If more than 1 study meets criteria, the study whose OR point estimate has the largest difference from the overall OR is selected for removal. Upon removal, MA is re-conducted, a second funnel plot is generated, and the process continues until the individual study’s OR CI overlap with the combined OR CI. Successful reduction of PB and SH was evaluated with I2 statistic. Results: Funnel plot analysis successfully reduced PB and SH from an I2 of 90% to an I2of 40%. Conclusion. Meta-analysis PB and SH can be reduced by graphical funnel plot method incorporating individual study confidence intervals.