A0716
Title: Statistical considerations for inverse publication bias in safety outcomes
Authors: Lifeng Lin - University of Arizona (United States) [presenting]
Abstract: Inverse publication bias (IPB) is an emerging concern in meta-analyses of safety outcomes, where studies favoring comparable safety profiles between interventions and controls may be more likely to be published. Unlike traditional publication bias, IPB is less recognized yet can lead to misleading conclusions in systematic reviews. Key statistical considerations are discussed for detecting IPB. First, methodological approaches are presented, including how contour-enhanced funnel plots and statistical tests such as Egger's and Peters' regressions can be adapted for IPB detection, emphasizing the impact of effect direction in bias assessments. Second, empirical analyses from meta-analyses of adverse events in the SMART Safety dataset, the largest evidence synthesis database for adverse events, are shared. Findings indicate that a considerable proportion of meta-analyses exhibit IPB and that qualitative assessments may be necessary to complement statistical methods, particularly in small-study contexts. The aim is to raise awareness of IPB and provide practical guidance for systematic reviews of safety outcomes.
