Journal of Statistics: Advances in Theory and ApplicationsAuthor's: Romain Piaget-Rossel and Patrick Taffé
Pages: [91] - [117]
Received Date: April 12, 2019; Revised May 3, 2019
Submitted by:
DOI: http://dx.doi.org/10.18642/jsata_7100122046
Mantel-Haenszel is a fixed-effect meta-analysis method, which performs quite well under the assumption of a homogeneous treatment effect, even in the presence of very rare events. However, this method fails to account for the information contained in single-arm and double zero studies. In this paper, we developed a pseudo-likelihood approach, which allows the inclusion of both single-arm and double-zero studies in the combined effect size estimate. Using Monte-Carlo simulations, we evaluated the behaviour of these two methods when subject to an increasing proportion of single-arm and double-zero studies. We found that the exclusion of double-zero studies did not impact the performance of the Mantel-Haenszel method, whereas the exclusion of single-arm studies reduced its efficiency compared to the pseudo-likelihood approach. We thus recommend using the pseudo-likelihood approach when the meta-analysis includes single-arm studies. With only double-zero studies, the Mantel-Haenszel can safely be used.
double-zero studies, Mantel-Haenszel, meta-analysis, Monte-Carlo simulations, pseudo-likelihood approach, single-arm studies.
