Related Experiment Videos
Bayesian methods for analysis of binary outcome data in cluster randomized trials on the absolute risk scale.
Simon G Thompson1, David E Warn, Rebecca M Turner
1MRC Biostatistics Unit, Institute of Public Health, Robinson Way, Cambridge CB2 2SR, U.K. simon.thompson@mrc-bsu.cam.ac.uk
Statistics in Medicine
|January 30, 2004
Summary
This study introduces a Bayesian hierarchical model for cluster randomized trials, enabling estimation on an absolute risk scale for clearer clinical interpretation. This approach enhances the analysis of binary outcomes and treatment effects in healthcare research.
Area of Science:
- Biostatistics
- Clinical Trials
- Bayesian Inference
Background:
- Bayesian hierarchical modeling offers advantages for cluster randomized trials, including parameter uncertainty and covariate flexibility.
- Previous methods for binary outcomes used log-odds scales, limiting clinical interpretability of treatment effects and intracluster correlation (ICC).
Purpose of the Study:
- To develop a Bayesian hierarchical model for cluster randomized trials that allows estimation on the absolute risk scale.
- To improve the clinical interpretability of treatment effects and between-cluster variance in binary outcome data analysis.
Main Methods:
- Developed a Bayesian hierarchical model for cluster randomized trials with binary outcomes.
- Enabled estimation on the absolute risk scale, moving beyond the log-odds ratio.
- Applied models to data from a coronary heart disease secondary prevention trial.
Main Results:
- The new model facilitates clinically interpretable estimation of treatment effects and between-cluster variance on the absolute risk scale.
- Demonstrated incorporation of prior data on ICCs and derivation of the number needed to treat.
- Showcased consideration of both cluster and individual-level covariates.
Conclusions:
- The developed Bayesian approach provides clinically meaningful results for cluster randomized trials with binary outcomes.
- This method enhances the utility of Bayesian modeling by offering interpretable estimates on the absolute risk scale.