Assessment of publication bias in meta-analyses of cardiovascular diseases

Abstract

Objective: To examine variables related with publication bias assessment in a sample of systematic reviews with meta-analysis on cardiovascular diseases.

Design: Systematic review of meta-analyses.

Setting: Journals indexed in Medline and the Cochrane Library.

Study population: 225 reviews with meta-analysis published between 1990 and 2002.

Data collection: Data from meta-analyses were gathered according to a structured protocol. The outcome was the assessment, not the existence, of publication bias by the original authors.

Results: Publication bias was assessed in 25 (11.1%) reviews, increasing with time: from 3.4% before 1998 to 19.0% in those published in 2002. A stepwise logistic regression model included several variables increasing the assessment of publication bias: number of primary studies (>7 compared with ≤7, odds ratio (OR)  =  5.40, 95% CI  =  1.36 to 21.44), number of searched databases (≥4 compared with <3, OR  =  8.58, 95% CI  =  1.73 to 42.62), to be a meta-analysis on observational studies (OR  =  3.60, 95% CI  =  1.04 to 12.49), and year of publication (2002 compared with <2000, OR  =  5.73, 95% CI  =  1.16 to 28.36). In reviews published in the Cochrane Library publication bias was less frequently assessed (OR  =  0.06, 95% CI  =  0.01 to 0.69).

Conclusions: The frequency of assessment of publication bias in meta-analysis is still very low, although it has improved with time. It is more frequent in meta-analyses on observational studies and it is related to other methodological characteristics of reviews.