Can authorship bias be detected in meta-analysis?

Abstract

Purpose: Statistical approaches have been developed to detect bias in individual trials, but guidance on how to detect systematic differences at a meta-analytical level is lacking. In this paper, we elucidate whether author bias can be detected in a cohort of randomized trials included in a meta-analysis. Methods: We utilized mortality data from 35 trials (10,880 patients) included in our previously published meta-analysis. First, we linked each author with their trial (or trials). Then we calculated author-specific odds ratios using univariate cross table methods. Finally, we tested the effect of authorship by comparing each author’s estimated odds ratio with all other pooled estimated odds ratios using meta-regression. Results: The median number of investigators named as authors on the primary trial reports was six (interquartile range: 5-8, range: 2-32). The results showed that the slope of author effect for mortality ranged from − 1.35 to 0.71. We identified only one author team showing a marginally significant effect (− 0.39; 95% confidence interval, − 0.78 to 0.00). This author team has a history of retractions due to data manipulations and ethical violations. Conclusion: When combining trial-level data to produce a pooled effect estimate, investigators must consider sources of potential bias. Our results suggest that systematic errors can be detected using meta-regression, although further research is needed to examine the sensitivity of this model. Systematic reviewers will benefit from the availability of methods to guard against the dissemination of results with the potential to mislead decision-making.