Comparing Methods for Analyzing Treatment Effects in Multicenter Trials
Author Information
Author(s): Chu Rong, Thabane Lehana, Ma Jinhui, Holbrook Anne, Pullenayegum Eleanor, Devereaux Philip James
Primary Institution: McMaster University
Hypothesis
What are the properties of commonly used statistical models for analyzing correlated continuous outcomes in multicenter randomized controlled trials?
Conclusion
All six models produced unbiased estimates of treatment effect, with the mixed-effects model being the most efficient.
Supporting Evidence
- All methods yielded unbiased estimates of treatment effect.
- Ignoring centers led to inflation of standard error and loss of statistical power.
- Mixed-effects model was most efficient across simulations.
- Fixed-effects model was less precise with a large number of centers.
- GEE underestimated standard error with a small number of centers.
- Center-level models had variable point estimates and low interval coverage.
Takeaway
This study looked at different ways to analyze data from trials with multiple centers, finding that all methods gave similar results, but some were better at estimating treatment effects.
Methodology
The study compared six statistical methods using simulations to evaluate their performance in estimating treatment effects in multicenter trials.
Limitations
The study did not address model performance when treatment by center interaction exists.
Participant Demographics
Patients were recruited from 46 family physician practices.
Statistical Information
Confidence Interval
(0.787, 1.753)
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website