©The Author(s) 2015. Published by Baishideng Publishing Group Inc. All rights reserved.
World J Meta-Anal. Dec 26, 2015; 3(6): 225-231
Published online Dec 26, 2015. doi: 10.13105/wjma.v3.i6.225
Published online Dec 26, 2015. doi: 10.13105/wjma.v3.i6.225
Towards better meta-analyses in assisted reproductive technology: Fixed, random or multivariate models?
Philippe Lehert, Faculty of Medicine, the University of Melbourne, Southbank 3006, Victoria, Australia
Philippe Lehert, Faculty of Economics, UCL Louvain University, B-7000 Mons, Belgium
Author contributions: This author is the exclusive author of this whole research.
Conflict-of-interest statement: The author declares no competing interests.
Data sharing statement: None.
Correspondence to: Dr. Philippe Lehert, PhD, Professor of Statistics, Faculty of Medicine, the University of Melbourne, 801/250 St Kilda Rd, Southbank 3006, Victoria, Australia. philippe.lehert@gmail.com
Telephone: +61-3-96999411 Fax: +61-3-96999411
Received: May 15, 2015
Peer-review started: May 20, 2015
First decision: July 26, 2015
Revised: September 27, 2015
Accepted: October 16, 2015
Article in press: October 19, 2015
Published online: December 26, 2015
Processing time: 223 Days and 6.9 Hours
Peer-review started: May 20, 2015
First decision: July 26, 2015
Revised: September 27, 2015
Accepted: October 16, 2015
Article in press: October 19, 2015
Published online: December 26, 2015
Processing time: 223 Days and 6.9 Hours
Core Tip
Core tip: The numerous meta-analyses (MA) published in assisted reproduction technology (ART) are often characterized by conflicting results. This paper provides evidence that the choice of the meta-analytical model constitutes a major concern. We first identified a general profile of characteristics of the ART studies, compare different models by simulation and resolve a practical case. MA based on the fixed model produce severe biases and falsely significant differences. Better results derive from the random model. For partially reported multiple endpoints, the multivariate model takes advantage of the between-endpoint inter-correlation and provides consistent estimates, better precision, and higher power.