Can authorship bias be detected in meta-analysis?

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Research Integrity

Background:

• Systematic reviews and meta-analyses synthesize evidence from multiple studies.
• Detecting bias at the individual trial level is established, but meta-analytical level bias detection is less understood.
• Author bias, a potential source of systematic error, can influence pooled results.

Purpose of the Study:

• To investigate the detectability of author bias within a cohort of randomized trials included in a meta-analysis.
• To determine if systematic differences attributable to authors can be identified at the meta-analytical level.

Main Methods:

• Utilized mortality data from 35 randomized trials (10,880 patients) from a prior meta-analysis.
• Linked authors to their respective trials and calculated author-specific odds ratios.
• Employed meta-regression to compare author-specific odds ratios against pooled estimates to detect systematic effects.

Main Results:

• Analyzed trials with a median of six authors.
• The slope of author effect for mortality varied widely (-1.35 to 0.71).
• One author team showed a marginally significant effect and had a history of retractions for data manipulation and ethical violations.

Conclusions:

• Meta-regression can potentially detect systematic errors arising from authorship in meta-analyses.
• Further research is needed to refine the sensitivity of these detection models.
• Methods to identify and mitigate author bias are crucial for systematic reviewers to prevent the dissemination of misleading findings.