Article
Beta-binomial and other models in meta-analyses with very few studies
Search Medline for
Authors
-
Moritz Felsch
- Institut für Qualität und Wirtschaftlichkeit im Gesundheitswesen (IQWiG), Köln, Germany
-
Lars Beckmann
- Institut für Qualität und Wirtschaftlichkeit im Gesundheitswesen (IQWiG), Köln, Germany
-
Ralf Bender
- Institut für Qualität und Wirtschaftlichkeit im Gesundheitswesen (IQWiG), Köln, Germany
-
Oliver Kuß
- Institut für Biometrie und Epidemiologie, Deutsches Diabetes-Zentrum (DDZ), Düsseldorf, Germany
-
Guido Skipka
- Institut für Qualität und Wirtschaftlichkeit im Gesundheitswesen (IQWiG), Köln, Germany
-
Tim Mathes
- Institut für Forschung in der Operativen Medizin (IFOM), Universität Witten/Herdecke, Köln, Germany
| Published: | February 26, 2021 |
|---|
Outline
Text
Background: Random effects meta-analyses face difficulties when used in situations with very few (2 – 4) studies [1]. As heterogeneity between the studies cannot be reliably estimated, this can lead to biased point estimates and too narrow confidence intervals [2]. Unfortunately, situations with very few situations are quite common [3].
Methods: In case of binary outcomes the standard (“common-rho”) beta-binomial model (BBM) has been proven to be a valuable statistical model in situations with very few studies or/and a low number of events [4], [5]. One drawback of this model is breaking of randomisation, in sense of ignoring the affiliation of the study arms to the same study. Therefore, we examined extensions of this model without breaking of randomisation and other alternative BBMs (“common-beta”). We compared these models with general linear mixed models (GLMMs), another major alternative in case of binary outcomes, and standard random effects meta-analyses models like Hartung-Knapp-Sidik-Jonkman (HKSJ) and DerSimonian-Lair (DSL) in a simulation study. For meta-analyses of 2, 3, 4, 5 and 10 studies, we generated 10000 data sets using empirical data of existing Cochrane Reviews [3] to simulate sample size of study, risk in control arm, heterogeneity between studies and true effect size of Odds Ratio.
Results: Bias was low for all methods. There was no difference in performance regarding bias, coverage probability and power between the “common-rho” BBM and the extension without breaking of randomisation in all scenarios (2, 3, 4, 5, 10 studies). “Common-rho” and “common-beta” BBMs performed quite similar, too. Coverage probabilities were above or at 95 % for HKSJ and BBMs in all scenarios. In case of 2 studies, all methods performed poorly with significance levels far above 5 % or power below 5 %. In case of 3, 4, 5 and 10 studies, BBMs had coverage probabilities closer to 95 % and greater but still low (<40 %) power compared to HKSJ.
Conclusion: In case of binary outcomes and very few studies, the BBM may be a valuable alternative to standard random effects meta-analyses models in certain situations.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
References
- 1.
- Veroniki AA, Jackson D, Viechtbauer W, Bender R, Bowden J, Knapp G, et al. Methods to estimate the between-study variance and its uncertainty in meta-analysis. Res Synth Methods. 2016;7(1):55-79.
- 2.
- Higgins JPT, Thompson SG, Spiegelhalter DJ. A re-evaluation of random-effects meta-analysis. J R Stat Soc Ser A Stat Soc. 2009;172(1):137-59.
- 3.
- Turner RM, Davey J, Clarke MJ, Thompson SG, Higgins JP. Predicting the extent of heterogeneity in meta-analysis, using empirical data from the Cochrane Database of Systematic Reviews. Int J Epidemiol. 2012;41(3):818-27.
- 4.
- Kuss O. Statistical methods for meta-analyses including information from studies without any events-add nothing to nothing and succeed nevertheless. Stat Med. 2015;34(7):1097-116.
- 5.
- Mathes T, Kuss O. A comparison of methods for meta-analysis of a small number of studies with binary outcomes. Res Synth Methods. 2018;9(3):366-81.
