BishtRef
Articles Journals Publishers Authors Subjects Most Cited New Papers Year Stats Open Access
journal article Dec 01, 2014

Statistical methods for meta‐analyses including information from studies without any events—add nothing to nothing and succeed nevertheless

O. Kuss
Statistics in Medicine Vol. 34 No. 7 pp. 1097-1116 · Wiley
View at Publisher Save 10.1002/sim.6383
Abstract
Meta‐analyses with rare events, especially those that include studies with no event in one (‘single‐zero’) or even both (‘double‐zero’) treatment arms, are still a statistical challenge. In the case of double‐zero studies, researchers in general delete these studies or use continuity corrections to avoid them. A number of arguments against both options has been given, and statistical methods that use the information from double‐zero studies without using continuity corrections have been proposed. In this paper, we collect them and compare them by simulation. This simulation study tries to mirror real‐life situations as completely as possible by deriving true underlying parameters from empirical data on actually performed meta‐analyses. It is shown that for each of the commonly encountered effect estimators valid statistical methods are available that use the information from double‐zero studies without using continuity corrections. Interestingly, all of them are truly random effects models, and so also the current standard method for very sparse data as recommended from the Cochrane collaboration, the Yusuf–Peto odds ratio, can be improved on. For actual analysis, we recommend to use beta‐binomial regression methods to arrive at summary estimates for the odds ratio, the relative risk, or the risk difference. Methods that ignore information from double‐zero studies or use continuity corrections should no longer be used. We illustrate the situation with an example where the original analysis ignores 35 double‐zero studies, and a superior analysis discovers a clinically relevant advantage of off‐pump surgery in coronary artery bypass grafting. Copyright © 2014 John Wiley & Sons, Ltd.
Topics

No keywords indexed for this article. Browse by subject →

References
57
[4]
KussO WandreyM KunzeM.How frequent are meta‐analyses with “double‐zero” studies in systematic reviews?2009. Available from:http://www.egms.de/static/de/meetings/gmds2009/09gmds155.shtml[Accessed 7 November 2014].
[11]
LiuD.Combining information for heterogeneous studies and rare events studies: a confidence distribution approach 2012. Available from:https://rucore.libraries.rutgers.edu/rutgers-lib/37435/[Accessed 7 November 2014].
[14]
Hirji KF (2006)
[22]
Kuss O "An exact test for meta‐analysis with binary endpoints" Methods of Information in Medicine (2007) 10.3414/me0422
[23]
Higgins JPT (2011)
[30]
McCulloch CE (2008)
[33]
Molenberghs G (2005)
[36]
Bias reduction of maximum likelihood estimates

DAVID FIRTH

Biometrika 10.1093/biomet/80.1.27
[40]
Sato T "On the variance estimator for the Mantel–Haenszel risk difference (letter)" Biometrics (1989)
[42]
Newman SC (2001)
[48]
Fan X (2002)
[49]
Sato T "Bias in the Peto one‐step estimator for the common odds ratio" Bulletin of Informatics and Cybernetics (2005) 10.5109/12587

Showing 50 of 57 references

Cited By
174
Random-effects meta-analysis models for pooling rare events data: a comparison between frequentist and bayesian methods

Minghong Yao, Ke Deng · 2025

BMC Medical Research Methodology
Beta-binomial models for meta-analysis with binary outcomes: Variations, extensions, and additional insights from econometrics

Tim Mathes, Oliver Kuss · 2021

Research Methods in Medicine &...
Safety Profile of Ceftazidime–Avibactam: Pooled Data from the Adult Phase II and Phase III Clinical Trial Programme

Karen Cheng, Paul Newell · 2020

Drug Safety
Pitfalls of using the risk ratio in meta‐analysis

Ilyas Bakbergenuly, David C. Hoaglin · 2019

Research Synthesis Methods
Meta-analysis of binary outcomes via generalized linear mixed models: a simulation study

Ilyas Bakbergenuly, Elena Kulinskaya · 2018

BMC Medical Research Methodology
Modelling multiple thresholds in meta-analysis of diagnostic test accuracy studies

Susanne Steinhauser, Martin Schumacher · 2016

BMC Medical Research Methodology
Methods to estimate the between‐study variance and its uncertainty in meta‐analysis

Areti Angeliki Veroniki, Dan Jackson · 2015

Research Synthesis Methods
Misunderstandings about Q and ‘Cochran's Q test' in meta‐analysis

David C. Hoaglin · 2015

Statistics in Medicine
Metrics
174
Citations
57
References
Details
Published
Dec 01, 2014
Vol/Issue
34(7)
Pages
1097-1116
License
View
Authors
O
O. Kuss

Institute for Biometry and Epidemiology German Diabetes Center, Leibniz Institute for Diabetes Research at Heinrich Heine University Düsseldorf Düsseldorf Germany

Funding
Deutsche Forschungsgemeinschaft Award: KU 1443/3-1
Cite This Article
O. Kuss (2014). Statistical methods for meta‐analyses including information from studies without any events—add nothing to nothing and succeed nevertheless. Statistics in Medicine, 34(7), 1097-1116. https://doi.org/10.1002/sim.6383
Related

You May Also Like

Quantifying heterogeneity in a meta‐analysis

Julian P. T. Higgins, Simon G. Thompson · 2002

31,148 citations

Multiple imputation using chained equations: Issues and guidance for practice

Ian R. White, Patrick Royston · 2010

7,694 citations

MULTIVARIABLE PROGNOSTIC MODELS: ISSUES IN DEVELOPING MODELS, EVALUATING ASSUMPTIONS AND ADEQUACY, AND MEASURING AND REDUCING ERRORS

Frank E. Harrell, Kerry L. Lee · 1996

7,135 citations

Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity‐score matched samples

Peter C. Austin · 2009

5,504 citations

Evaluating the added predictive ability of a new marker: From area under the ROC curve to reclassification and beyond

Michael J. Pencina, Ralph B. D' Agostino · 2007

5,289 citations