Artikel
Bivariate meta-analysis with insufficient reporting of the correlation between outcomes on the study level
Suche in Medline nach
Autoren
-
Christian Röver
- Department of Medical Statistics, University Medical Center Göttingen, Göttingen, Germany
-
Tim Friede
- Department of Medical Statistics, University Medical Center Göttingen, Göttingen, Germany
| Veröffentlicht: | 6. September 2024 |
|---|
Gliederung
Text
The joint consideration of several (correlated) endpoints in a meta-analysis promises advantages in terms of a gain in precision [1]. However, bivariate methods are rarely used, possibly because explicit information on the correlation is not commonly available from study reports. We aimed to still take advantage of correlated outcomes by treating correlations as unknowns and estimating these from the data between studies rather than within studies. The generic (univariate) random-effects meta-analysis model is generalized to include either common or random (within-study) correlations, as well as correlated between-study heterogeneity. We implemented the models in a Bayesian framework. Using some examples we investigate their gains in precision for overall effects or shrinkage estimates over univariate methods. Of particular interest may be shrinkage estimates in cases where only an estimate of one of the two endpoints is available. For comparison, we also consider (a Bayesian version of) the bivariate model proposed by Riley et al. [2].
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
References
- 1.
- van Houwelingen HC, Zwinderman KH, Stijnen T. A bivariate approach to meta-analysis. Statistics in Medicine. 1993;12(24):2273–2284.
- 2.
- Riley RD, Thompson JR, Abrams KR. An alternative model for bivariate random-effects meta-analysis when the within-study correlations are unknown. Biostatistics. 2008;9(1):172–186.
