A Bayesian model averaging approach with non-informative priors for cost-effectiveness analyses

Area of Science:

• Health economics
• Biostatistics
• Clinical trial analysis

Background:

• Assessing technology cost-effectiveness relies on clinical trial data for costs and effects.
• Skewed and heavy-tailed cost data distributions complicate realistic probabilistic modeling.

Purpose of the Study:

• To develop a robust method for cost-effectiveness analysis (CEA) integrating model uncertainty.
• To address challenges posed by non-standard cost data distributions in CEA.

Main Methods:

• Utilizing Bayesian model averaging (BMA) for cost-effectiveness acceptability curves (CEACs).
• Employing fractional Bayes factors to handle undetermined marginal densities in BMA.
• Implementing path sampling for efficient computation of integral ratios.

Main Results:

• The proposed Bayesian model averaging approach provides a framework for handling complex cost data.
• Fractional Bayes factors offer a viable solution for undetermined marginal densities in BMA for CEA.
• Path sampling enables efficient computation, making the method practical.

Conclusions:

• The Bayesian model averaging method enhances the reliability of cost-effectiveness assessments, especially with challenging cost data.
• This approach improves the integration of model uncertainty into economic evaluations.
• The proposed methodology offers a more realistic assessment of new and existing technologies' value.