Bayesian hierarchical methods for meta-analysis combining randomized-controlled and single-arm studies.

Area of Science:

• Biostatistics
• Evidence Synthesis
• Clinical Trial Methodology

Background:

• Meta-analysis traditionally relies on randomized controlled trials (RCTs).
• Synthesizing evidence from single-arm studies alongside RCTs is crucial when RCT generalizability is limited or single-arm data dominates.
• Single-arm studies present challenges due to potential selection bias and confounding factors, impacting reliability.

Purpose of the Study:

• To propose a Bayesian hierarchical framework for bias reduction and efficiency gain in meta-analysis.
• To develop and evaluate methods for synthesizing evidence from both randomized controlled trials and single-arm studies.
• To offer flexible downweighting of single-arm studies within the framework.

Main Methods:

• A Bayesian hierarchical framework incorporating three methods: bivariate generalized linear mixed effects models, hierarchical power prior model, and hierarchical commensurate prior model.
• Consideration of design differences and potential biases in all proposed models.
• Flexible downweighting of single-arm studies using hierarchical power prior and hierarchical commensurate prior models.

Main Results:

• The proposed Bayesian hierarchical framework effectively addresses bias and enhances efficiency in evidence synthesis.
• Simulation studies and application to motivating datasets demonstrate the performance of the developed methods.
• The hierarchical commensurate prior model shows particular accuracy and robustness for evidence synthesis.

Conclusions:

• The Bayesian hierarchical framework provides a robust approach for integrating diverse evidence sources in meta-analysis.
• The hierarchical commensurate prior model is recommended as the primary method due to its superior performance.
• The methods offer valuable tools for improving the reliability and generalizability of synthesized evidence.