Related Experiment Videos
Meta-analysis of multiple outcomes by regression with random effects
C S Berkey1, D C Hoaglin, A Antczak-Bouckoms
1Department of Medicine, Brigham and Women's Hospital, Boston, MA, USA. catherine.berkey@channing.harvard.edu
Statistics in Medicine
|December 5, 1998
Summary
New random-effects models improve regression meta-analysis for multiple correlated outcomes, addressing unexplained heterogeneity and potential bias found in fixed-effects models.
Area of Science:
- Biostatistics
- Clinical Trial Analysis
- Epidemiology
Background:
- Fixed-effects meta-analysis can jointly analyze multiple correlated outcomes but may be biased by unexplained heterogeneity.
- Existing methods may not adequately account for residual variation among studies after covariate adjustment.
Purpose of the Study:
- To propose and evaluate novel random-effects regression models for meta-analysis of multiple correlated outcomes.
- To compare the performance of these new models against fixed-effects and separate-outcomes approaches.
Main Methods:
- Developed two random-effects regression models for handling multiple, correlated outcomes in meta-analysis.
- Compared proposed models with fixed-effects generalized-least-squares and separate-outcomes models.
- Utilized a simulation study and a meta-analysis of periodontal clinical trials for evaluation.
Main Results:
- Random-effects models demonstrated advantages over fixed-effects and separate-outcomes models in simulation studies.
- The proposed methods effectively handle unexplained heterogeneity in regression meta-analysis.
- New approaches facilitate meta-analysis of trials with more than two treatment arms.
Conclusions:
- Random-effects regression meta-analysis is a superior approach for multiple correlated outcomes when unexplained heterogeneity exists.
- These methods offer a more robust and less biased alternative to fixed-effects models.
- The proposed techniques enhance the analysis of complex clinical trial data and comparative effectiveness research.