Victor M Lu1,2, Kevin Phan3, Julia X M Yin4, Kerrie L McDonald4,3. 1. Cure Brain Cancer Neuro-oncology Laboratory, Prince of Wales Clinical School, Lowy Cancer Research Centre, University of New South Wales, Sydney, Australia. victor.lu@student.unsw.edu.au. 2. Prince of Wales Clinical School, University of New South Wales, Randwick, Sydney, NSW, 2031, Australia. victor.lu@student.unsw.edu.au. 3. Prince of Wales Clinical School, University of New South Wales, Randwick, Sydney, NSW, 2031, Australia. 4. Cure Brain Cancer Neuro-oncology Laboratory, Prince of Wales Clinical School, Lowy Cancer Research Centre, University of New South Wales, Sydney, Australia.
Abstract
INTRODUCTION: There are many potential biomarkers in glioblastoma (GBM), and meta-analyses represent the highest level of evidence when inferring their prognostic significance. It is possible however, that inherent design properties of the studies included in these meta-analyses may affect the pooled hazard ratio (HR) of the meta-analyses. This meta-epidemiological study aims to investigate the potential bias of three study-level properties in meta-analyses of GBM biomarkers currently published in the literature. METHODS: Seven electronic databases from inception to December 2017 were searched for meta-analyses evaluating different GBM biomarkers, which were screened against specific criteria. Study-level data were extracted from each meta-analysis, and analyzed using logistic regression to yield the ratio of HR (RHR) summary statistic. RESULTS: Nine meta-analyses investigating different GBM biomarkers were included. Of all the meta-analyses, the HRs of two studies were significantly underestimated by older studies; they investigated biomarkers IDH1 (RHR = 1.145; p = 0.017) and CD133 (RHR = 0.850; p = 0.013). Study-level size and design showed non-significant trends towards affecting the overall HR in all included meta-analyses. CONCLUSIONS: This meta-epidemiological study demonstrated that study-level year can already significantly affect the pooled HR of GBM biomarkers reported by meta-analyses. It is possible that in the future, more study-level properties will exert significant effect. In terms of future GBM biomarker meta-analyses, special consideration of bias should be given to these study-level properties as potential sources of effect on the prognostic pooled HR.
INTRODUCTION: There are many potential biomarkers in glioblastoma (GBM), and meta-analyses represent the highest level of evidence when inferring their prognostic significance. It is possible however, that inherent design properties of the studies included in these meta-analyses may affect the pooled hazard ratio (HR) of the meta-analyses. This meta-epidemiological study aims to investigate the potential bias of three study-level properties in meta-analyses of GBM biomarkers currently published in the literature. METHODS: Seven electronic databases from inception to December 2017 were searched for meta-analyses evaluating different GBM biomarkers, which were screened against specific criteria. Study-level data were extracted from each meta-analysis, and analyzed using logistic regression to yield the ratio of HR (RHR) summary statistic. RESULTS: Nine meta-analyses investigating different GBM biomarkers were included. Of all the meta-analyses, the HRs of two studies were significantly underestimated by older studies; they investigated biomarkers IDH1 (RHR = 1.145; p = 0.017) and CD133 (RHR = 0.850; p = 0.013). Study-level size and design showed non-significant trends towards affecting the overall HR in all included meta-analyses. CONCLUSIONS: This meta-epidemiological study demonstrated that study-level year can already significantly affect the pooled HR of GBM biomarkers reported by meta-analyses. It is possible that in the future, more study-level properties will exert significant effect. In terms of future GBM biomarker meta-analyses, special consideration of bias should be given to these study-level properties as potential sources of effect on the prognostic pooled HR.
Authors: A D Oxman; G H Guyatt; J Singer; C H Goldsmith; B G Hutchison; R A Milner; D L Streiner Journal: J Clin Epidemiol Date: 1991 Impact factor: 6.437
Authors: Jonathan A C Sterne; Alex J Sutton; John P A Ioannidis; Norma Terrin; David R Jones; Joseph Lau; James Carpenter; Gerta Rücker; Roger M Harbord; Christopher H Schmid; Jennifer Tetzlaff; Jonathan J Deeks; Jaime Peters; Petra Macaskill; Guido Schwarzer; Sue Duval; Douglas G Altman; David Moher; Julian P T Higgins Journal: BMJ Date: 2011-07-22
Authors: Jelena Savović; Ross J Harris; Lesley Wood; Rebecca Beynon; Doug Altman; Bodil Als-Nielsen; Ethan M Balk; Jonathan Deeks; Lise Lotte Gluud; Christian Gluud; John P A Ioannidis; Peter Jűni; David Moher; Julie Pildal; Kenneth F Schulz; Jonathan A C Sterne Journal: Res Synth Methods Date: 2010-11-30 Impact factor: 5.273