Publication bias in hematology systematic reviews and meta-analyses.

Background: PB occurs when research yielding statistically significant results are published at a higher rate than studies producing statistically non-significant results
Objective: Our primary objective was to evaluate the manners in which hematology authors attempt to limit PB in their systematic reviews (SRs). Our secondary objective was to analyze for PB in the reviews that did not already perform an evaluation.
Methods: We searched PubMed for SRs and meta-analyses published between 2011 and 2015 in the top 9 hematology journals by Impact Factor. For secondary analysis, we evaluated for PB using Trim and Fill, Cumulative Meta-Analysis by Precision, and Selection Models.
Results: We found that 59.4% (79/133) of included records mentioned PB and 49.6% (66/133) included and assessment for PB. The most common method of assessment for PB was visual inspection of a funnel plot (48/66, 72.7%). The majority of the included SRs did not report adherence to any reporting guideline (89/133, 66.9%). One hundred (75.2%) SRs manually searched the reference lists of their included studies, while only 8 (6.0%) manually searched journals. Gray literature was searched in 61 SRs (45.9%). Trim and Fill, cumulative meta-analysis by precision, and selection models detected PB 60%, 90%, and 0% of the time, respectively.
Discussion: There is room for improvement in rates of discussion and evaluation of PB in hematology SRs. Additionally, methods by which authors can mitigate PB, such as searching gray literature, were infrequently included in our sample. When we evaluated for PB in non-reporting SRs we detected PB frequently, with varied rates of detection across different tests. This indicates the need for multiple methods of assessment to ensure proper estimation of PB present in SRs.