Related Experiment Videos
Meta-analysis combining parallel and cross-over clinical trials. I: Continuous outcomes
François Curtin1, Douglas G Altman, Diana Elbourne
1Medical Statistics Unit, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1 7HT, UK. francoiscurtin@compuserve.com
Statistics in Medicine
|September 5, 2002
Summary
This study examines pooling continuous outcomes from parallel and cross-over trials using weighted mean difference (WMD) and standardized weighted mean difference (SWMD) methods in meta-analysis. Cross-over designs have a significant weight, potentially introducing design-specific bias.
Area of Science:
- Biostatistics
- Clinical Trials Methodology
- Evidence Synthesis
Background:
- Clinical trials frequently employ both parallel and cross-over designs for treatment assessment.
- Combining these designs in meta-analysis requires specific statistical approaches.
Purpose of the Study:
- To explore methods for pooling continuous outcomes in meta-analyses that include both parallel and cross-over trial designs.
- To compare the weighted mean difference (WMD) and standardized weighted mean difference (SWMD) approaches.
Main Methods:
- Meta-analysis formulae were developed for combined parallel and cross-over designs.
- A weighted average of treatment estimates from both designs was utilized.
- Random effects models were considered for implementation.
- The study evaluated both WMD and SWMD for pooling continuous outcomes.
Main Results:
- The relative weight of cross-over designs can be substantial in combined-design meta-analyses, irrespective of the number of subjects.
- Differences in weight estimation between WMD and SWMD can further amplify the influence of cross-over trials.
- This amplification highlights a potential for design-specific bias.
Conclusions:
- Both WMD and SWMD are viable methods for meta-analysis of combined parallel and cross-over designs, with the choice dependent on outcome type.
- The significant weight of cross-over designs necessitates careful consideration to mitigate potential design-specific biases in meta-analysis.